initial split of C files to be used both by Matlab and Python

the regularizers routines have been splitted to separate the Matlab from
the C implementation. This will allow the concurrent use of the C code
from Boost Python.

15 files changed, 1619 insertions, 1091 deletions

diff --git a/main_func/regularizers_CPU/FGP_TV.c b/main_func/regularizers_CPU/FGP_TV.c
index 1a1fd13..5d8cfb9 100644
--- a/main_func/regularizers_CPU/FGP_TV.c
+++ b/main_func/regularizers_CPU/FGP_TV.c
@@ -1,10 +1,23 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
 #include "mex.h"
-#include <matrix.h>
-#include <math.h>
-#include <stdlib.h>
-#include <memory.h>
-#include <stdio.h>
-#include "omp.h"
+#include "FGP_TV_core.h"
 
 /* C-OMP implementation of FGP-TV [1] denoising/regularization model (2D/3D case)
  *
@@ -33,17 +46,6 @@
  *
  */
 
-float copyIm(float *A, float *B, int dimX, int dimY, int dimZ);
-float Obj_func2D(float *A, float *D, float *R1, float *R2, float lambda, int dimX, int dimY);
-float Grad_func2D(float *P1, float *P2, float *D, float *R1, float *R2, float lambda, int dimX, int dimY);
-float Proj_func2D(float *P1, float *P2, int methTV, int dimX, int dimY);
-float Rupd_func2D(float *P1, float *P1_old, float *P2, float *P2_old, float *R1, float *R2, float tkp1, float tk, int dimX, int dimY);
-
-float Obj_func3D(float *A, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ);
-float Grad_func3D(float *P1, float *P2, float *P3, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ);
-float Proj_func3D(float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ);
-float Rupd_func3D(float *P1, float *P1_old, float *P2, float *P2_old, float *P3, float *P3_old, float *R1, float *R2, float *R3, float tkp1, float tk, int dimX, int dimY, int dimZ);
-
 
 void mexFunction(
         int nlhs, mxArray *plhs[],
@@ -236,165 +238,3 @@ void mexFunction(
         printf("FGP-TV iterations stopped at iteration %i with the function value %f \n", ll, funcvalA[0]);
     }
 }
-
-/* 2D-case related Functions */
-/*****************************************************************/
-float Obj_func2D(float *A, float *D, float *R1, float *R2, float lambda, int dimX, int dimY)
-{
-    float val1, val2;
-    int i,j;
-#pragma omp parallel for shared(A,D,R1,R2) private(i,j,val1,val2)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* boundary conditions  */
-            if (i == 0) {val1 = 0.0f;} else {val1 = R1[(i-1)*dimY + (j)];}
-            if (j == 0) {val2 = 0.0f;} else {val2 = R2[(i)*dimY + (j-1)];}
-            D[(i)*dimY + (j)] = A[(i)*dimY + (j)] - lambda*(R1[(i)*dimY + (j)] + R2[(i)*dimY + (j)] - val1 - val2);
-        }}
-    return *D;
-}
-float Grad_func2D(float *P1, float *P2, float *D, float *R1, float *R2, float lambda, int dimX, int dimY)
-{
-    float val1, val2, multip;
-    int i,j;
-    multip = (1.0f/(8.0f*lambda));
-#pragma omp parallel for shared(P1,P2,D,R1,R2,multip) private(i,j,val1,val2)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* boundary conditions */
-            if (i == dimX-1) val1 = 0.0f; else val1 = D[(i)*dimY + (j)] - D[(i+1)*dimY + (j)];
-            if (j == dimY-1) val2 = 0.0f; else val2 = D[(i)*dimY + (j)] - D[(i)*dimY + (j+1)];
-            P1[(i)*dimY + (j)] = R1[(i)*dimY + (j)] + multip*val1;
-            P2[(i)*dimY + (j)] = R2[(i)*dimY + (j)] + multip*val2;
-        }}
-    return 1;
-}
-float Proj_func2D(float *P1, float *P2, int methTV, int dimX, int dimY)
-{
-    float val1, val2, denom;
-    int i,j;
-    if (methTV == 0) {
-        /* isotropic TV*/
-#pragma omp parallel for shared(P1,P2) private(i,j,denom)
-        for(i=0; i<dimX; i++) {
-            for(j=0; j<dimY; j++) {
-                denom = pow(P1[(i)*dimY + (j)],2) +  pow(P2[(i)*dimY + (j)],2);
-                if (denom > 1) {
-                    P1[(i)*dimY + (j)] = P1[(i)*dimY + (j)]/sqrt(denom);
-                    P2[(i)*dimY + (j)] = P2[(i)*dimY + (j)]/sqrt(denom);
-                }
-            }}
-    }
-    else {
-        /* anisotropic TV*/
-#pragma omp parallel for shared(P1,P2) private(i,j,val1,val2)
-        for(i=0; i<dimX; i++) {
-            for(j=0; j<dimY; j++) {
-                val1 = fabs(P1[(i)*dimY + (j)]);
-                val2 = fabs(P2[(i)*dimY + (j)]);
-                if (val1 < 1.0f) {val1 = 1.0f;}
-                if (val2 < 1.0f) {val2 = 1.0f;}
-                P1[(i)*dimY + (j)] = P1[(i)*dimY + (j)]/val1;
-                P2[(i)*dimY + (j)] = P2[(i)*dimY + (j)]/val2;
-            }}
-    }
-    return 1;
-}
-float Rupd_func2D(float *P1, float *P1_old, float *P2, float *P2_old, float *R1, float *R2, float tkp1, float tk, int dimX, int dimY)
-{
-    int i,j;
-    float multip;
-    multip = ((tk-1.0f)/tkp1);
-#pragma omp parallel for shared(P1,P2,P1_old,P2_old,R1,R2,multip) private(i,j)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            R1[(i)*dimY + (j)] = P1[(i)*dimY + (j)] + multip*(P1[(i)*dimY + (j)] - P1_old[(i)*dimY + (j)]);
-            R2[(i)*dimY + (j)] = P2[(i)*dimY + (j)] + multip*(P2[(i)*dimY + (j)] - P2_old[(i)*dimY + (j)]);
-        }}
-    return 1;
-}
-
-/* 3D-case related Functions */
-/*****************************************************************/
-float Obj_func3D(float *A, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ)
-{
-    float val1, val2, val3;
-    int i,j,k;
-#pragma omp parallel for shared(A,D,R1,R2,R3) private(i,j,k,val1,val2,val3)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                /* boundary conditions */
-                if (i == 0) {val1 = 0.0f;} else {val1 = R1[(dimX*dimY)*k + (i-1)*dimY + (j)];}
-                if (j == 0) {val2 = 0.0f;} else {val2 = R2[(dimX*dimY)*k + (i)*dimY + (j-1)];}
-                if (k == 0) {val3 = 0.0f;} else {val3 = R3[(dimX*dimY)*(k-1) + (i)*dimY + (j)];}
-                D[(dimX*dimY)*k + (i)*dimY + (j)] = A[(dimX*dimY)*k + (i)*dimY + (j)] - lambda*(R1[(dimX*dimY)*k + (i)*dimY + (j)] + R2[(dimX*dimY)*k + (i)*dimY + (j)] + R3[(dimX*dimY)*k + (i)*dimY + (j)] - val1 - val2 - val3);
-            }}}
-    return *D;
-}
-float Grad_func3D(float *P1, float *P2, float *P3, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ)
-{
-    float val1, val2, val3, multip;
-    int i,j,k;
-    multip = (1.0f/(8.0f*lambda));
-#pragma omp parallel for shared(P1,P2,P3,D,R1,R2,R3,multip) private(i,j,k,val1,val2,val3)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                /* boundary conditions */
-                if (i == dimX-1) val1 = 0.0f; else val1 = D[(dimX*dimY)*k + (i)*dimY + (j)] - D[(dimX*dimY)*k + (i+1)*dimY + (j)];
-                if (j == dimY-1) val2 = 0.0f; else val2 = D[(dimX*dimY)*k + (i)*dimY + (j)] - D[(dimX*dimY)*k + (i)*dimY + (j+1)];
-                if (k == dimZ-1) val3 = 0.0f; else val3 = D[(dimX*dimY)*k + (i)*dimY + (j)] - D[(dimX*dimY)*(k+1) + (i)*dimY + (j)];
-                P1[(dimX*dimY)*k + (i)*dimY + (j)] = R1[(dimX*dimY)*k + (i)*dimY + (j)] + multip*val1;
-                P2[(dimX*dimY)*k + (i)*dimY + (j)] = R2[(dimX*dimY)*k + (i)*dimY + (j)] + multip*val2;
-                P3[(dimX*dimY)*k + (i)*dimY + (j)] = R3[(dimX*dimY)*k + (i)*dimY + (j)] + multip*val3;
-            }}}
-    return 1;
-}
-float Proj_func3D(float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ)
-{
-    float val1, val2, val3;
-    int i,j,k;
-#pragma omp parallel for shared(P1,P2,P3) private(i,j,k,val1,val2,val3)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                val1 = fabs(P1[(dimX*dimY)*k + (i)*dimY + (j)]);
-                val2 = fabs(P2[(dimX*dimY)*k + (i)*dimY + (j)]);
-                val3 = fabs(P3[(dimX*dimY)*k + (i)*dimY + (j)]);
-                if (val1 < 1.0f) {val1 = 1.0f;}
-                if (val2 < 1.0f) {val2 = 1.0f;}
-                if (val3 < 1.0f) {val3 = 1.0f;}
-                
-                P1[(dimX*dimY)*k + (i)*dimY + (j)] = P1[(dimX*dimY)*k + (i)*dimY + (j)]/val1;
-                P2[(dimX*dimY)*k + (i)*dimY + (j)] = P2[(dimX*dimY)*k + (i)*dimY + (j)]/val2;
-                P3[(dimX*dimY)*k + (i)*dimY + (j)] = P3[(dimX*dimY)*k + (i)*dimY + (j)]/val3;
-            }}}
-    return 1;
-}
-float Rupd_func3D(float *P1, float *P1_old, float *P2, float *P2_old, float *P3, float *P3_old, float *R1, float *R2, float *R3, float tkp1, float tk, int dimX, int dimY, int dimZ)
-{
-    int i,j,k;
-    float multip;
-    multip = ((tk-1.0f)/tkp1);
-#pragma omp parallel for shared(P1,P2,P3,P1_old,P2_old,P3_old,R1,R2,R3,multip) private(i,j,k)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                R1[(dimX*dimY)*k + (i)*dimY + (j)] = P1[(dimX*dimY)*k + (i)*dimY + (j)] + multip*(P1[(dimX*dimY)*k + (i)*dimY + (j)] - P1_old[(dimX*dimY)*k + (i)*dimY + (j)]);
-                R2[(dimX*dimY)*k + (i)*dimY + (j)] = P2[(dimX*dimY)*k + (i)*dimY + (j)] + multip*(P2[(dimX*dimY)*k + (i)*dimY + (j)] - P2_old[(dimX*dimY)*k + (i)*dimY + (j)]);
-                R3[(dimX*dimY)*k + (i)*dimY + (j)] = P3[(dimX*dimY)*k + (i)*dimY + (j)] + multip*(P3[(dimX*dimY)*k + (i)*dimY + (j)] - P3_old[(dimX*dimY)*k + (i)*dimY + (j)]);
-            }}}
-    return 1;
-}
-
-/* General Functions */
-/*****************************************************************/
-/* Copy Image */
-float copyIm(float *A, float *B, int dimX, int dimY, int dimZ)
-{
-    int j;
-#pragma omp parallel for shared(A, B) private(j)
-    for(j=0; j<dimX*dimY*dimZ; j++)  B[j] = A[j];
-    return *B;
-}
diff --git a/main_func/regularizers_CPU/FGP_TV_core.c b/main_func/regularizers_CPU/FGP_TV_core.c
new file mode 100644
index 0000000..c383a71
--- /dev/null
+++ b/main_func/regularizers_CPU/FGP_TV_core.c
@@ -0,0 +1,200 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include "FGP_TV_core.h"
+
+/* 2D-case related Functions */
+/*****************************************************************/
+float Obj_func2D(float *A, float *D, float *R1, float *R2, float lambda, int dimX, int dimY)
+{
+	float val1, val2;
+	int i, j;
+#pragma omp parallel for shared(A,D,R1,R2) private(i,j,val1,val2)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			/* boundary conditions  */
+			if (i == 0) { val1 = 0.0f; }
+			else { val1 = R1[(i - 1)*dimY + (j)]; }
+			if (j == 0) { val2 = 0.0f; }
+			else { val2 = R2[(i)*dimY + (j - 1)]; }
+			D[(i)*dimY + (j)] = A[(i)*dimY + (j)] - lambda*(R1[(i)*dimY + (j)] + R2[(i)*dimY + (j)] - val1 - val2);
+		}
+	}
+	return *D;
+}
+float Grad_func2D(float *P1, float *P2, float *D, float *R1, float *R2, float lambda, int dimX, int dimY)
+{
+	float val1, val2, multip;
+	int i, j;
+	multip = (1.0f / (8.0f*lambda));
+#pragma omp parallel for shared(P1,P2,D,R1,R2,multip) private(i,j,val1,val2)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			/* boundary conditions */
+			if (i == dimX - 1) val1 = 0.0f; else val1 = D[(i)*dimY + (j)] - D[(i + 1)*dimY + (j)];
+			if (j == dimY - 1) val2 = 0.0f; else val2 = D[(i)*dimY + (j)] - D[(i)*dimY + (j + 1)];
+			P1[(i)*dimY + (j)] = R1[(i)*dimY + (j)] + multip*val1;
+			P2[(i)*dimY + (j)] = R2[(i)*dimY + (j)] + multip*val2;
+		}
+	}
+	return 1;
+}
+float Proj_func2D(float *P1, float *P2, int methTV, int dimX, int dimY)
+{
+	float val1, val2, denom;
+	int i, j;
+	if (methTV == 0) {
+		/* isotropic TV*/
+#pragma omp parallel for shared(P1,P2) private(i,j,denom)
+		for (i = 0; i<dimX; i++) {
+			for (j = 0; j<dimY; j++) {
+				denom = pow(P1[(i)*dimY + (j)], 2) + pow(P2[(i)*dimY + (j)], 2);
+				if (denom > 1) {
+					P1[(i)*dimY + (j)] = P1[(i)*dimY + (j)] / sqrt(denom);
+					P2[(i)*dimY + (j)] = P2[(i)*dimY + (j)] / sqrt(denom);
+				}
+			}
+		}
+	}
+	else {
+		/* anisotropic TV*/
+#pragma omp parallel for shared(P1,P2) private(i,j,val1,val2)
+		for (i = 0; i<dimX; i++) {
+			for (j = 0; j<dimY; j++) {
+				val1 = fabs(P1[(i)*dimY + (j)]);
+				val2 = fabs(P2[(i)*dimY + (j)]);
+				if (val1 < 1.0f) { val1 = 1.0f; }
+				if (val2 < 1.0f) { val2 = 1.0f; }
+				P1[(i)*dimY + (j)] = P1[(i)*dimY + (j)] / val1;
+				P2[(i)*dimY + (j)] = P2[(i)*dimY + (j)] / val2;
+			}
+		}
+	}
+	return 1;
+}
+float Rupd_func2D(float *P1, float *P1_old, float *P2, float *P2_old, float *R1, float *R2, float tkp1, float tk, int dimX, int dimY)
+{
+	int i, j;
+	float multip;
+	multip = ((tk - 1.0f) / tkp1);
+#pragma omp parallel for shared(P1,P2,P1_old,P2_old,R1,R2,multip) private(i,j)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			R1[(i)*dimY + (j)] = P1[(i)*dimY + (j)] + multip*(P1[(i)*dimY + (j)] - P1_old[(i)*dimY + (j)]);
+			R2[(i)*dimY + (j)] = P2[(i)*dimY + (j)] + multip*(P2[(i)*dimY + (j)] - P2_old[(i)*dimY + (j)]);
+		}
+	}
+	return 1;
+}
+
+/* 3D-case related Functions */
+/*****************************************************************/
+float Obj_func3D(float *A, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ)
+{
+	float val1, val2, val3;
+	int i, j, k;
+#pragma omp parallel for shared(A,D,R1,R2,R3) private(i,j,k,val1,val2,val3)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+				/* boundary conditions */
+				if (i == 0) { val1 = 0.0f; }
+				else { val1 = R1[(dimX*dimY)*k + (i - 1)*dimY + (j)]; }
+				if (j == 0) { val2 = 0.0f; }
+				else { val2 = R2[(dimX*dimY)*k + (i)*dimY + (j - 1)]; }
+				if (k == 0) { val3 = 0.0f; }
+				else { val3 = R3[(dimX*dimY)*(k - 1) + (i)*dimY + (j)]; }
+				D[(dimX*dimY)*k + (i)*dimY + (j)] = A[(dimX*dimY)*k + (i)*dimY + (j)] - lambda*(R1[(dimX*dimY)*k + (i)*dimY + (j)] + R2[(dimX*dimY)*k + (i)*dimY + (j)] + R3[(dimX*dimY)*k + (i)*dimY + (j)] - val1 - val2 - val3);
+			}
+		}
+	}
+	return *D;
+}
+float Grad_func3D(float *P1, float *P2, float *P3, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ)
+{
+	float val1, val2, val3, multip;
+	int i, j, k;
+	multip = (1.0f / (8.0f*lambda));
+#pragma omp parallel for shared(P1,P2,P3,D,R1,R2,R3,multip) private(i,j,k,val1,val2,val3)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+				/* boundary conditions */
+				if (i == dimX - 1) val1 = 0.0f; else val1 = D[(dimX*dimY)*k + (i)*dimY + (j)] - D[(dimX*dimY)*k + (i + 1)*dimY + (j)];
+				if (j == dimY - 1) val2 = 0.0f; else val2 = D[(dimX*dimY)*k + (i)*dimY + (j)] - D[(dimX*dimY)*k + (i)*dimY + (j + 1)];
+				if (k == dimZ - 1) val3 = 0.0f; else val3 = D[(dimX*dimY)*k + (i)*dimY + (j)] - D[(dimX*dimY)*(k + 1) + (i)*dimY + (j)];
+				P1[(dimX*dimY)*k + (i)*dimY + (j)] = R1[(dimX*dimY)*k + (i)*dimY + (j)] + multip*val1;
+				P2[(dimX*dimY)*k + (i)*dimY + (j)] = R2[(dimX*dimY)*k + (i)*dimY + (j)] + multip*val2;
+				P3[(dimX*dimY)*k + (i)*dimY + (j)] = R3[(dimX*dimY)*k + (i)*dimY + (j)] + multip*val3;
+			}
+		}
+	}
+	return 1;
+}
+float Proj_func3D(float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ)
+{
+	float val1, val2, val3;
+	int i, j, k;
+#pragma omp parallel for shared(P1,P2,P3) private(i,j,k,val1,val2,val3)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+				val1 = fabs(P1[(dimX*dimY)*k + (i)*dimY + (j)]);
+				val2 = fabs(P2[(dimX*dimY)*k + (i)*dimY + (j)]);
+				val3 = fabs(P3[(dimX*dimY)*k + (i)*dimY + (j)]);
+				if (val1 < 1.0f) { val1 = 1.0f; }
+				if (val2 < 1.0f) { val2 = 1.0f; }
+				if (val3 < 1.0f) { val3 = 1.0f; }
+
+				P1[(dimX*dimY)*k + (i)*dimY + (j)] = P1[(dimX*dimY)*k + (i)*dimY + (j)] / val1;
+				P2[(dimX*dimY)*k + (i)*dimY + (j)] = P2[(dimX*dimY)*k + (i)*dimY + (j)] / val2;
+				P3[(dimX*dimY)*k + (i)*dimY + (j)] = P3[(dimX*dimY)*k + (i)*dimY + (j)] / val3;
+			}
+		}
+	}
+	return 1;
+}
+float Rupd_func3D(float *P1, float *P1_old, float *P2, float *P2_old, float *P3, float *P3_old, float *R1, float *R2, float *R3, float tkp1, float tk, int dimX, int dimY, int dimZ)
+{
+	int i, j, k;
+	float multip;
+	multip = ((tk - 1.0f) / tkp1);
+#pragma omp parallel for shared(P1,P2,P3,P1_old,P2_old,P3_old,R1,R2,R3,multip) private(i,j,k)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+				R1[(dimX*dimY)*k + (i)*dimY + (j)] = P1[(dimX*dimY)*k + (i)*dimY + (j)] + multip*(P1[(dimX*dimY)*k + (i)*dimY + (j)] - P1_old[(dimX*dimY)*k + (i)*dimY + (j)]);
+				R2[(dimX*dimY)*k + (i)*dimY + (j)] = P2[(dimX*dimY)*k + (i)*dimY + (j)] + multip*(P2[(dimX*dimY)*k + (i)*dimY + (j)] - P2_old[(dimX*dimY)*k + (i)*dimY + (j)]);
+				R3[(dimX*dimY)*k + (i)*dimY + (j)] = P3[(dimX*dimY)*k + (i)*dimY + (j)] + multip*(P3[(dimX*dimY)*k + (i)*dimY + (j)] - P3_old[(dimX*dimY)*k + (i)*dimY + (j)]);
+			}
+		}
+	}
+	return 1;
+}
+
+/* General Functions */
+/*****************************************************************/
+/* Copy Image */
+float copyIm(float *A, float *B, int dimX, int dimY, int dimZ)
+{
+	int j;
+#pragma omp parallel for shared(A, B) private(j)
+	for (j = 0; j<dimX*dimY*dimZ; j++)  B[j] = A[j];
+	return *B;
+}
diff --git a/main_func/regularizers_CPU/FGP_TV_core.h b/main_func/regularizers_CPU/FGP_TV_core.h
new file mode 100644
index 0000000..8d6af3e
--- /dev/null
+++ b/main_func/regularizers_CPU/FGP_TV_core.h
@@ -0,0 +1,63 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include <matrix.h>
+#include <math.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <stdio.h>
+#include "omp.h"
+
+/* C-OMP implementation of FGP-TV [1] denoising/regularization model (2D/3D case)
+*
+* Input Parameters:
+* 1. Noisy image/volume [REQUIRED]
+* 2. lambda - regularization parameter [REQUIRED]
+* 3. Number of iterations [OPTIONAL parameter]
+* 4. eplsilon: tolerance constant [OPTIONAL parameter]
+* 5. TV-type: 'iso' or 'l1' [OPTIONAL parameter]
+*
+* Output:
+* [1] Filtered/regularized image
+* [2] last function value
+*
+* Example of image denoising:
+* figure;
+* Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+* u0 = Im + .05*randn(size(Im)); % adding noise
+* u = FGP_TV(single(u0), 0.05, 100, 1e-04);
+*
+* to compile with OMP support: mex FGP_TV.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
+* This function is based on the Matlab's code and paper by
+* [1] Amir Beck and Marc Teboulle, "Fast Gradient-Based Algorithms for Constrained Total Variation Image Denoising and Deblurring Problems"
+*
+* D. Kazantsev, 2016-17
+*
+*/
+
+float copyIm(float *A, float *B, int dimX, int dimY, int dimZ);
+float Obj_func2D(float *A, float *D, float *R1, float *R2, float lambda, int dimX, int dimY);
+float Grad_func2D(float *P1, float *P2, float *D, float *R1, float *R2, float lambda, int dimX, int dimY);
+float Proj_func2D(float *P1, float *P2, int methTV, int dimX, int dimY);
+float Rupd_func2D(float *P1, float *P1_old, float *P2, float *P2_old, float *R1, float *R2, float tkp1, float tk, int dimX, int dimY);
+
+float Obj_func3D(float *A, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ);
+float Grad_func3D(float *P1, float *P2, float *P3, float *D, float *R1, float *R2, float *R3, float lambda, int dimX, int dimY, int dimZ);
+float Proj_func3D(float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ);
+float Rupd_func3D(float *P1, float *P1_old, float *P2, float *P2_old, float *P3, float *P3_old, float *R1, float *R2, float *R3, float tkp1, float tk, int dimX, int dimY, int dimZ);
diff --git a/main_func/regularizers_CPU/LLT_model.c b/main_func/regularizers_CPU/LLT_model.c
index 0aed31e..6b50d33 100644
--- a/main_func/regularizers_CPU/LLT_model.c
+++ b/main_func/regularizers_CPU/LLT_model.c
@@ -1,49 +1,25 @@
-#include "mex.h"
-#include <matrix.h>
-#include <math.h>
-#include <stdlib.h>
-#include <memory.h>
-#include <stdio.h>
-#include "omp.h"
-
-#define EPS 0.01
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
 
-/* C-OMP implementation of Lysaker, Lundervold and Tai (LLT) model of higher order regularization penalty
- *
- * Input Parameters:
- * 1. U0 - origanal noise image/volume
- * 2. lambda - regularization parameter
- * 3. tau - time-step  for explicit scheme 
- * 4. iter - iterations number
- * 5. epsil  - tolerance constant (to terminate earlier) 
- * 6. switcher - default is 0, switch to (1) to restrictive smoothing in Z dimension (in test)
- *
- * Output:
- * Filtered/regularized image
- *
- * Example:
- * figure;
- * Im = double(imread('lena_gray_256.tif'))/255;  % loading image
- * u0 = Im + .03*randn(size(Im)); % adding noise
- * [Den] = LLT_model(single(u0), 10, 0.1, 1);
- *
- *
- * to compile with OMP support: mex LLT_model.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
- * References: Lysaker, Lundervold and Tai (LLT) 2003, IEEE
- *
- * 28.11.16/Harwell
- */
-/* 2D functions */
-float der2D(float *U, float *D1, float *D2, int dimX, int dimY, int dimZ);
-float div_upd2D(float *U0, float *U, float *D1, float *D2, int dimX, int dimY, int dimZ, float lambda, float tau);
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
 
-float der3D(float *U, float *D1, float *D2, float *D3, int dimX, int dimY, int dimZ);
-float div_upd3D(float *U0, float *U, float *D1, float *D2, float *D3,  unsigned short *Map, int switcher, int dimX, int dimY, int dimZ, float lambda, float tau);
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
 
-float calcMap(float *U, unsigned short *Map, int dimX, int dimY, int dimZ);
-float cleanMap(unsigned short *Map, int dimX, int dimY, int dimZ);
+#include "mex.h"
+#include "LLT_model_core.h"
 
-float copyIm(float *A, float *U, int dimX, int dimY, int dimZ);
 
 void mexFunction(
         int nlhs, mxArray *plhs[],
@@ -165,267 +141,3 @@ void mexFunction(
         printf("HO iterations stopped at iteration: %i\n", ll);       
     }
 }
-
-float der2D(float *U, float *D1, float *D2, int dimX, int dimY, int dimZ)
-{
-    int i, j, i_p, i_m, j_m, j_p;
-    float dxx, dyy, denom_xx, denom_yy;
-#pragma omp parallel for shared(U,D1,D2) private(i, j, i_p, i_m, j_m, j_p, denom_xx, denom_yy, dxx, dyy)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            i_p = i + 1; if (i_p == dimX) i_p = i - 1;
-            i_m = i - 1; if (i_m < 0) i_m = i + 1;
-            j_p = j + 1; if (j_p == dimY) j_p = j - 1;
-            j_m = j - 1; if (j_m < 0) j_m = j + 1;
-            
-            dxx = U[i_p*dimY + j] - 2.0f*U[i*dimY + j] + U[i_m*dimY + j];
-            dyy = U[i*dimY + j_p] - 2.0f*U[i*dimY + j] + U[i*dimY + j_m];
-                        
-            denom_xx = fabs(dxx) + EPS;
-            denom_yy = fabs(dyy) + EPS;
-            
-            D1[i*dimY + j] = dxx/denom_xx;
-            D2[i*dimY + j] = dyy/denom_yy;
-        }}
-    return 1;
-}
-float div_upd2D(float *U0, float *U, float *D1, float *D2, int dimX, int dimY, int dimZ, float lambda, float tau)
-{
-    int i, j, i_p, i_m, j_m, j_p;
-    float div, dxx, dyy;
-#pragma omp parallel for shared(U,U0,D1,D2) private(i, j, i_p, i_m, j_m, j_p, div, dxx, dyy)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            i_p = i + 1; if (i_p == dimX) i_p = i - 1;
-            i_m = i - 1; if (i_m < 0) i_m = i + 1;
-            j_p = j + 1; if (j_p == dimY) j_p = j - 1;
-            j_m = j - 1; if (j_m < 0) j_m = j + 1;
-            
-            dxx = D1[i_p*dimY + j] - 2.0f*D1[i*dimY + j] + D1[i_m*dimY + j];
-            dyy = D2[i*dimY + j_p] - 2.0f*D2[i*dimY + j] + D2[i*dimY + j_m];
-            
-            div = dxx + dyy;
-            
-            U[i*dimY + j] = U[i*dimY + j] - tau*div - tau*lambda*(U[i*dimY + j] - U0[i*dimY + j]);
-        }}
-    return *U0;
-}
-
-float der3D(float *U, float *D1, float *D2, float *D3, int dimX, int dimY, int dimZ)
-{
-    int i, j, k, i_p, i_m, j_m, j_p, k_p, k_m;
-    float dxx, dyy, dzz, denom_xx, denom_yy, denom_zz;
-#pragma omp parallel for shared(U,D1,D2,D3) private(i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, denom_xx, denom_yy, denom_zz, dxx, dyy, dzz)
-    for(i=0; i<dimX; i++) {
-        /* symmetric boundary conditions (Neuman) */
-        i_p = i + 1; if (i_p == dimX) i_p = i - 1;
-        i_m = i - 1; if (i_m < 0) i_m = i + 1;
-        for(j=0; j<dimY; j++) {
-            j_p = j + 1; if (j_p == dimY) j_p = j - 1;
-            j_m = j - 1; if (j_m < 0) j_m = j + 1;
-            for(k=0; k<dimZ; k++) {
-                k_p = k + 1; if (k_p == dimZ) k_p = k - 1;
-                k_m = k - 1; if (k_m < 0) k_m = k + 1;
-                
-                dxx = U[dimX*dimY*k + i_p*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i_m*dimY + j];
-                dyy = U[dimX*dimY*k + i*dimY + j_p] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i*dimY + j_m];
-                dzz = U[dimX*dimY*k_p + i*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k_m + i*dimY + j];                
-                
-                denom_xx = fabs(dxx) + EPS;
-                denom_yy = fabs(dyy) + EPS;
-                denom_zz = fabs(dzz) + EPS;
-                
-                D1[dimX*dimY*k + i*dimY + j] = dxx/denom_xx;
-                D2[dimX*dimY*k + i*dimY + j] = dyy/denom_yy;
-                D3[dimX*dimY*k + i*dimY + j] = dzz/denom_zz;               
-                
-            }}}
-    return 1;
-}
-
-float div_upd3D(float *U0, float *U, float *D1, float *D2, float *D3,  unsigned short *Map, int switcher, int dimX, int dimY, int dimZ, float lambda, float tau)
-{
-    int i, j, k, i_p, i_m, j_m, j_p, k_p, k_m;
-    float div, dxx, dyy, dzz;
-#pragma omp parallel for shared(U,U0,D1,D2,D3) private(i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, div, dxx, dyy, dzz)
-    for(i=0; i<dimX; i++) {
-        /* symmetric boundary conditions (Neuman) */
-        i_p = i + 1; if (i_p == dimX) i_p = i - 1;
-        i_m = i - 1; if (i_m < 0) i_m = i + 1;
-        for(j=0; j<dimY; j++) {
-            j_p = j + 1; if (j_p == dimY) j_p = j - 1;
-            j_m = j - 1; if (j_m < 0) j_m = j + 1;
-            for(k=0; k<dimZ; k++) {
-                k_p = k + 1; if (k_p == dimZ) k_p = k - 1;
-                k_m = k - 1; if (k_m < 0) k_m = k + 1;
-//                 k_p1 = k + 2; if (k_p1 >= dimZ) k_p1 = k - 2;
-//                 k_m1 = k - 2; if (k_m1 < 0) k_m1 = k + 2;   
-                
-                dxx = D1[dimX*dimY*k + i_p*dimY + j] - 2.0f*D1[dimX*dimY*k + i*dimY + j] + D1[dimX*dimY*k + i_m*dimY + j];
-                dyy = D2[dimX*dimY*k + i*dimY + j_p] - 2.0f*D2[dimX*dimY*k + i*dimY + j] + D2[dimX*dimY*k + i*dimY + j_m];               
-                dzz = D3[dimX*dimY*k_p + i*dimY + j] - 2.0f*D3[dimX*dimY*k + i*dimY + j] + D3[dimX*dimY*k_m + i*dimY + j];
-                
-                if ((switcher == 1) && (Map[dimX*dimY*k + i*dimY + j] == 0)) dzz = 0;                
-                div = dxx + dyy + dzz;
-                
-//                 if (switcher == 1) {                    
-                    // if (Map2[dimX*dimY*k + i*dimY + j] == 0) dzz2 = 0;
-                    //else dzz2 = D4[dimX*dimY*k_p1 + i*dimY + j] - 2.0f*D4[dimX*dimY*k + i*dimY + j] + D4[dimX*dimY*k_m1 + i*dimY + j];
-//                     div = dzz + dzz2;
-//                 }
-                  
-//                 dzz = D3[dimX*dimY*k_p + i*dimY + j] - 2.0f*D3[dimX*dimY*k + i*dimY + j] + D3[dimX*dimY*k_m + i*dimY + j];
-//                 dzz2 = D4[dimX*dimY*k_p1 + i*dimY + j] - 2.0f*D4[dimX*dimY*k + i*dimY + j] + D4[dimX*dimY*k_m1 + i*dimY + j];  
-//                 div = dzz + dzz2;
-                                
-                U[dimX*dimY*k + i*dimY + j] = U[dimX*dimY*k + i*dimY + j] - tau*div - tau*lambda*(U[dimX*dimY*k + i*dimY + j] - U0[dimX*dimY*k + i*dimY + j]);
-            }}}
-        return *U0;
- }   
-
-// float der3D_2(float *U, float *D1, float *D2, float *D3, float *D4, int dimX, int dimY, int dimZ)
-// {
-//     int i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, k_p1, k_m1;
-//     float dxx, dyy, dzz, dzz2, denom_xx, denom_yy, denom_zz, denom_zz2;
-// #pragma omp parallel for shared(U,D1,D2,D3,D4) private(i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, denom_xx, denom_yy, denom_zz, denom_zz2, dxx, dyy, dzz, dzz2, k_p1, k_m1)
-//     for(i=0; i<dimX; i++) {
-//         /* symmetric boundary conditions (Neuman) */
-//         i_p = i + 1; if (i_p == dimX) i_p = i - 1;
-//         i_m = i - 1; if (i_m < 0) i_m = i + 1;
-//         for(j=0; j<dimY; j++) {
-//             j_p = j + 1; if (j_p == dimY) j_p = j - 1;
-//             j_m = j - 1; if (j_m < 0) j_m = j + 1;
-//             for(k=0; k<dimZ; k++) {
-//                 k_p = k + 1; if (k_p == dimZ) k_p = k - 1;
-//                 k_m = k - 1; if (k_m < 0) k_m = k + 1;
-//                 k_p1 = k + 2; if (k_p1 >= dimZ) k_p1 = k - 2;
-//                 k_m1 = k - 2; if (k_m1 < 0) k_m1 = k + 2;                
-//                 
-//                 dxx = U[dimX*dimY*k + i_p*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i_m*dimY + j];
-//                 dyy = U[dimX*dimY*k + i*dimY + j_p] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i*dimY + j_m];
-//                 dzz = U[dimX*dimY*k_p + i*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k_m + i*dimY + j];                
-//                 dzz2 = U[dimX*dimY*k_p1 + i*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k_m1 + i*dimY + j];                
-//                 
-//                 denom_xx = fabs(dxx) + EPS;
-//                 denom_yy = fabs(dyy) + EPS;
-//                 denom_zz = fabs(dzz) + EPS;
-//                 denom_zz2 = fabs(dzz2) + EPS;
-//                 
-//                 D1[dimX*dimY*k + i*dimY + j] = dxx/denom_xx;
-//                 D2[dimX*dimY*k + i*dimY + j] = dyy/denom_yy;
-//                 D3[dimX*dimY*k + i*dimY + j] = dzz/denom_zz;               
-//                 D4[dimX*dimY*k + i*dimY + j] = dzz2/denom_zz2;                               
-//             }}}
-//     return 1;
-// }
-
-float calcMap(float *U, unsigned short *Map,  int dimX, int dimY, int dimZ)
-{  
-    int i,j,k,i1,j1,i2,j2,windowSize;
-    float val1, val2,thresh_val,maxval; 
-    windowSize = 1;
-    thresh_val = 0.0001; /*thresh_val = 0.0035;*/
-    
-    /* normalize volume first */
-    maxval = 0.0f;
-     for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {  
-             for(k=0; k<dimZ; k++) {  
-                if (U[dimX*dimY*k + i*dimY + j] > maxval) maxval = U[dimX*dimY*k + i*dimY + j];
-             }}}
-    
-    if (maxval != 0.0f) {
-     for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {  
-             for(k=0; k<dimZ; k++) {  
-               U[dimX*dimY*k + i*dimY + j] = U[dimX*dimY*k + i*dimY + j]/maxval;
-             }}}
-    }
-    else {
-       printf("%s \n", "Maximum value is zero!");       
-       return 0;
-    }
-    
-    #pragma omp parallel for shared(U,Map) private(i, j, k, i1, j1, i2, j2, val1, val2)
-     for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {  
-             for(k=0; k<dimZ; k++) {         
-                
-                Map[dimX*dimY*k + i*dimY + j] = 0; 
-//                 Map2[dimX*dimY*k + i*dimY + j] = 0; 
-                
-                val1 = 0.0f; val2 = 0.0f; 
-                for(i1=-windowSize; i1<=windowSize; i1++) {
-                    for(j1=-windowSize; j1<=windowSize; j1++) {
-                    i2 = i+i1;
-                    j2 = j+j1;
-                    
-                    if ((i2 >= 0) && (i2 < dimX) && (j2 >= 0) && (j2 < dimY)) {
-                      if (k == 0) {
-                          val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+1) + i2*dimY + j2],2);                        
-//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+2) + i2*dimY + j2],2);                                                  
-                      }
-                      else if (k == dimZ-1) {
-                          val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-1) + i2*dimY + j2],2); 
-//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-2) + i2*dimY + j2],2);                           
-                      }
-//                       else if (k == 1) {
-//                           val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-1) + i2*dimY + j2],2); 
-//                           val2 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+1) + i2*dimY + j2],2);  
-//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+2) + i2*dimY + j2],2);                           
-//                       }
-//                       else if (k == dimZ-2) {
-//                           val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-1) + i2*dimY + j2],2); 
-//                           val2 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+1) + i2*dimY + j2],2);      
-//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-2) + i2*dimY + j2],2);                           
-//                       }                      
-                      else {
-                          val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-1) + i2*dimY + j2],2); 
-                          val2 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+1) + i2*dimY + j2],2);                           
-//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-2) + i2*dimY + j2],2); 
-//                           val4 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+2) + i2*dimY + j2],2);  
-                      }
-                    }                    
-                    }}
-                
-                 val1 = 0.111f*val1; val2 = 0.111f*val2;
-//                  val3 = 0.111f*val3; val4 = 0.111f*val4;
-                 if ((val1 <= thresh_val) && (val2 <= thresh_val)) Map[dimX*dimY*k + i*dimY + j] = 1;                        
-//                  if ((val3 <= thresh_val) && (val4 <= thresh_val)) Map2[dimX*dimY*k + i*dimY + j] = 1;                        
-             }}}
-     return 1;
-}
-
-float cleanMap(unsigned short *Map, int dimX, int dimY, int dimZ)
-{  
-    int i, j, k, i1, j1, i2, j2, counter; 
-     #pragma omp parallel for shared(Map) private(i, j, k, i1, j1, i2, j2, counter)
-     for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {  
-             for(k=0; k<dimZ; k++) {   
-                    
-                 counter=0;
-                 for(i1=-3; i1<=3; i1++) {
-                    for(j1=-3; j1<=3; j1++) {
-                    i2 = i+i1;
-                    j2 = j+j1;
-                    if ((i2 >= 0) && (i2 < dimX) && (j2 >= 0) && (j2 < dimY)) {
-                    if (Map[dimX*dimY*k + i2*dimY + j2] == 0) counter++;                                       
-                    }
-                    }}
-                 if (counter < 24) Map[dimX*dimY*k + i*dimY + j] = 1;
-             }}}
-    return *Map;
-}
-
-    /* Copy Image */
-    float copyIm(float *A, float *U, int dimX, int dimY, int dimZ)
-    {
-        int j;
-#pragma omp parallel for shared(A, U) private(j)
-        for(j=0; j<dimX*dimY*dimZ; j++)  U[j] = A[j];
-        return *U;
-    }
-    /*********************3D *********************/
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/LLT_model_core.c b/main_func/regularizers_CPU/LLT_model_core.c
new file mode 100644
index 0000000..3098269
--- /dev/null
+++ b/main_func/regularizers_CPU/LLT_model_core.c
@@ -0,0 +1,300 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include "LLT_model_core.h"
+
+float der2D(float *U, float *D1, float *D2, int dimX, int dimY, int dimZ)
+{
+	int i, j, i_p, i_m, j_m, j_p;
+	float dxx, dyy, denom_xx, denom_yy;
+#pragma omp parallel for shared(U,D1,D2) private(i, j, i_p, i_m, j_m, j_p, denom_xx, denom_yy, dxx, dyy)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			/* symmetric boundary conditions (Neuman) */
+			i_p = i + 1; if (i_p == dimX) i_p = i - 1;
+			i_m = i - 1; if (i_m < 0) i_m = i + 1;
+			j_p = j + 1; if (j_p == dimY) j_p = j - 1;
+			j_m = j - 1; if (j_m < 0) j_m = j + 1;
+
+			dxx = U[i_p*dimY + j] - 2.0f*U[i*dimY + j] + U[i_m*dimY + j];
+			dyy = U[i*dimY + j_p] - 2.0f*U[i*dimY + j] + U[i*dimY + j_m];
+
+			denom_xx = fabs(dxx) + EPS;
+			denom_yy = fabs(dyy) + EPS;
+
+			D1[i*dimY + j] = dxx / denom_xx;
+			D2[i*dimY + j] = dyy / denom_yy;
+		}
+	}
+	return 1;
+}
+float div_upd2D(float *U0, float *U, float *D1, float *D2, int dimX, int dimY, int dimZ, float lambda, float tau)
+{
+	int i, j, i_p, i_m, j_m, j_p;
+	float div, dxx, dyy;
+#pragma omp parallel for shared(U,U0,D1,D2) private(i, j, i_p, i_m, j_m, j_p, div, dxx, dyy)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			/* symmetric boundary conditions (Neuman) */
+			i_p = i + 1; if (i_p == dimX) i_p = i - 1;
+			i_m = i - 1; if (i_m < 0) i_m = i + 1;
+			j_p = j + 1; if (j_p == dimY) j_p = j - 1;
+			j_m = j - 1; if (j_m < 0) j_m = j + 1;
+
+			dxx = D1[i_p*dimY + j] - 2.0f*D1[i*dimY + j] + D1[i_m*dimY + j];
+			dyy = D2[i*dimY + j_p] - 2.0f*D2[i*dimY + j] + D2[i*dimY + j_m];
+
+			div = dxx + dyy;
+
+			U[i*dimY + j] = U[i*dimY + j] - tau*div - tau*lambda*(U[i*dimY + j] - U0[i*dimY + j]);
+		}
+	}
+	return *U0;
+}
+
+float der3D(float *U, float *D1, float *D2, float *D3, int dimX, int dimY, int dimZ)
+{
+	int i, j, k, i_p, i_m, j_m, j_p, k_p, k_m;
+	float dxx, dyy, dzz, denom_xx, denom_yy, denom_zz;
+#pragma omp parallel for shared(U,D1,D2,D3) private(i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, denom_xx, denom_yy, denom_zz, dxx, dyy, dzz)
+	for (i = 0; i<dimX; i++) {
+		/* symmetric boundary conditions (Neuman) */
+		i_p = i + 1; if (i_p == dimX) i_p = i - 1;
+		i_m = i - 1; if (i_m < 0) i_m = i + 1;
+		for (j = 0; j<dimY; j++) {
+			j_p = j + 1; if (j_p == dimY) j_p = j - 1;
+			j_m = j - 1; if (j_m < 0) j_m = j + 1;
+			for (k = 0; k<dimZ; k++) {
+				k_p = k + 1; if (k_p == dimZ) k_p = k - 1;
+				k_m = k - 1; if (k_m < 0) k_m = k + 1;
+
+				dxx = U[dimX*dimY*k + i_p*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i_m*dimY + j];
+				dyy = U[dimX*dimY*k + i*dimY + j_p] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i*dimY + j_m];
+				dzz = U[dimX*dimY*k_p + i*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k_m + i*dimY + j];
+
+				denom_xx = fabs(dxx) + EPS;
+				denom_yy = fabs(dyy) + EPS;
+				denom_zz = fabs(dzz) + EPS;
+
+				D1[dimX*dimY*k + i*dimY + j] = dxx / denom_xx;
+				D2[dimX*dimY*k + i*dimY + j] = dyy / denom_yy;
+				D3[dimX*dimY*k + i*dimY + j] = dzz / denom_zz;
+
+			}
+		}
+	}
+	return 1;
+}
+
+float div_upd3D(float *U0, float *U, float *D1, float *D2, float *D3, unsigned short *Map, int switcher, int dimX, int dimY, int dimZ, float lambda, float tau)
+{
+	int i, j, k, i_p, i_m, j_m, j_p, k_p, k_m;
+	float div, dxx, dyy, dzz;
+#pragma omp parallel for shared(U,U0,D1,D2,D3) private(i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, div, dxx, dyy, dzz)
+	for (i = 0; i<dimX; i++) {
+		/* symmetric boundary conditions (Neuman) */
+		i_p = i + 1; if (i_p == dimX) i_p = i - 1;
+		i_m = i - 1; if (i_m < 0) i_m = i + 1;
+		for (j = 0; j<dimY; j++) {
+			j_p = j + 1; if (j_p == dimY) j_p = j - 1;
+			j_m = j - 1; if (j_m < 0) j_m = j + 1;
+			for (k = 0; k<dimZ; k++) {
+				k_p = k + 1; if (k_p == dimZ) k_p = k - 1;
+				k_m = k - 1; if (k_m < 0) k_m = k + 1;
+				//                 k_p1 = k + 2; if (k_p1 >= dimZ) k_p1 = k - 2;
+				//                 k_m1 = k - 2; if (k_m1 < 0) k_m1 = k + 2;   
+
+				dxx = D1[dimX*dimY*k + i_p*dimY + j] - 2.0f*D1[dimX*dimY*k + i*dimY + j] + D1[dimX*dimY*k + i_m*dimY + j];
+				dyy = D2[dimX*dimY*k + i*dimY + j_p] - 2.0f*D2[dimX*dimY*k + i*dimY + j] + D2[dimX*dimY*k + i*dimY + j_m];
+				dzz = D3[dimX*dimY*k_p + i*dimY + j] - 2.0f*D3[dimX*dimY*k + i*dimY + j] + D3[dimX*dimY*k_m + i*dimY + j];
+
+				if ((switcher == 1) && (Map[dimX*dimY*k + i*dimY + j] == 0)) dzz = 0;
+				div = dxx + dyy + dzz;
+
+				//                 if (switcher == 1) {                    
+				// if (Map2[dimX*dimY*k + i*dimY + j] == 0) dzz2 = 0;
+				//else dzz2 = D4[dimX*dimY*k_p1 + i*dimY + j] - 2.0f*D4[dimX*dimY*k + i*dimY + j] + D4[dimX*dimY*k_m1 + i*dimY + j];
+				//                     div = dzz + dzz2;
+				//                 }
+
+				//                 dzz = D3[dimX*dimY*k_p + i*dimY + j] - 2.0f*D3[dimX*dimY*k + i*dimY + j] + D3[dimX*dimY*k_m + i*dimY + j];
+				//                 dzz2 = D4[dimX*dimY*k_p1 + i*dimY + j] - 2.0f*D4[dimX*dimY*k + i*dimY + j] + D4[dimX*dimY*k_m1 + i*dimY + j];  
+				//                 div = dzz + dzz2;
+
+				U[dimX*dimY*k + i*dimY + j] = U[dimX*dimY*k + i*dimY + j] - tau*div - tau*lambda*(U[dimX*dimY*k + i*dimY + j] - U0[dimX*dimY*k + i*dimY + j]);
+			}
+		}
+	}
+	return *U0;
+}
+
+// float der3D_2(float *U, float *D1, float *D2, float *D3, float *D4, int dimX, int dimY, int dimZ)
+// {
+//     int i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, k_p1, k_m1;
+//     float dxx, dyy, dzz, dzz2, denom_xx, denom_yy, denom_zz, denom_zz2;
+// #pragma omp parallel for shared(U,D1,D2,D3,D4) private(i, j, k, i_p, i_m, j_m, j_p, k_p, k_m, denom_xx, denom_yy, denom_zz, denom_zz2, dxx, dyy, dzz, dzz2, k_p1, k_m1)
+//     for(i=0; i<dimX; i++) {
+//         /* symmetric boundary conditions (Neuman) */
+//         i_p = i + 1; if (i_p == dimX) i_p = i - 1;
+//         i_m = i - 1; if (i_m < 0) i_m = i + 1;
+//         for(j=0; j<dimY; j++) {
+//             j_p = j + 1; if (j_p == dimY) j_p = j - 1;
+//             j_m = j - 1; if (j_m < 0) j_m = j + 1;
+//             for(k=0; k<dimZ; k++) {
+//                 k_p = k + 1; if (k_p == dimZ) k_p = k - 1;
+//                 k_m = k - 1; if (k_m < 0) k_m = k + 1;
+//                 k_p1 = k + 2; if (k_p1 >= dimZ) k_p1 = k - 2;
+//                 k_m1 = k - 2; if (k_m1 < 0) k_m1 = k + 2;                
+//                 
+//                 dxx = U[dimX*dimY*k + i_p*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i_m*dimY + j];
+//                 dyy = U[dimX*dimY*k + i*dimY + j_p] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k + i*dimY + j_m];
+//                 dzz = U[dimX*dimY*k_p + i*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k_m + i*dimY + j];                
+//                 dzz2 = U[dimX*dimY*k_p1 + i*dimY + j] - 2.0f*U[dimX*dimY*k + i*dimY + j] + U[dimX*dimY*k_m1 + i*dimY + j];                
+//                 
+//                 denom_xx = fabs(dxx) + EPS;
+//                 denom_yy = fabs(dyy) + EPS;
+//                 denom_zz = fabs(dzz) + EPS;
+//                 denom_zz2 = fabs(dzz2) + EPS;
+//                 
+//                 D1[dimX*dimY*k + i*dimY + j] = dxx/denom_xx;
+//                 D2[dimX*dimY*k + i*dimY + j] = dyy/denom_yy;
+//                 D3[dimX*dimY*k + i*dimY + j] = dzz/denom_zz;               
+//                 D4[dimX*dimY*k + i*dimY + j] = dzz2/denom_zz2;                               
+//             }}}
+//     return 1;
+// }
+
+float calcMap(float *U, unsigned short *Map, int dimX, int dimY, int dimZ)
+{
+	int i, j, k, i1, j1, i2, j2, windowSize;
+	float val1, val2, thresh_val, maxval;
+	windowSize = 1;
+	thresh_val = 0.0001; /*thresh_val = 0.0035;*/
+
+						 /* normalize volume first */
+	maxval = 0.0f;
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+				if (U[dimX*dimY*k + i*dimY + j] > maxval) maxval = U[dimX*dimY*k + i*dimY + j];
+			}
+		}
+	}
+
+	if (maxval != 0.0f) {
+		for (i = 0; i<dimX; i++) {
+			for (j = 0; j<dimY; j++) {
+				for (k = 0; k<dimZ; k++) {
+					U[dimX*dimY*k + i*dimY + j] = U[dimX*dimY*k + i*dimY + j] / maxval;
+				}
+			}
+		}
+	}
+	else {
+		printf("%s \n", "Maximum value is zero!");
+		return 0;
+	}
+
+#pragma omp parallel for shared(U,Map) private(i, j, k, i1, j1, i2, j2, val1, val2)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+
+				Map[dimX*dimY*k + i*dimY + j] = 0;
+				//                 Map2[dimX*dimY*k + i*dimY + j] = 0; 
+
+				val1 = 0.0f; val2 = 0.0f;
+				for (i1 = -windowSize; i1 <= windowSize; i1++) {
+					for (j1 = -windowSize; j1 <= windowSize; j1++) {
+						i2 = i + i1;
+						j2 = j + j1;
+
+						if ((i2 >= 0) && (i2 < dimX) && (j2 >= 0) && (j2 < dimY)) {
+							if (k == 0) {
+								val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k + 1) + i2*dimY + j2], 2);
+								//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+2) + i2*dimY + j2],2);                                                  
+							}
+							else if (k == dimZ - 1) {
+								val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k - 1) + i2*dimY + j2], 2);
+								//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-2) + i2*dimY + j2],2);                           
+							}
+							//                       else if (k == 1) {
+							//                           val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-1) + i2*dimY + j2],2); 
+							//                           val2 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+1) + i2*dimY + j2],2);  
+							//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+2) + i2*dimY + j2],2);                           
+							//                       }
+							//                       else if (k == dimZ-2) {
+							//                           val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-1) + i2*dimY + j2],2); 
+							//                           val2 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+1) + i2*dimY + j2],2);      
+							//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-2) + i2*dimY + j2],2);                           
+							//                       }                      
+							else {
+								val1 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k - 1) + i2*dimY + j2], 2);
+								val2 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k + 1) + i2*dimY + j2], 2);
+								//                           val3 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k-2) + i2*dimY + j2],2); 
+								//                           val4 += pow(U[dimX*dimY*k + i2*dimY + j2] - U[dimX*dimY*(k+2) + i2*dimY + j2],2);  
+							}
+						}
+					}
+				}
+
+				val1 = 0.111f*val1; val2 = 0.111f*val2;
+				//                  val3 = 0.111f*val3; val4 = 0.111f*val4;
+				if ((val1 <= thresh_val) && (val2 <= thresh_val)) Map[dimX*dimY*k + i*dimY + j] = 1;
+				//                  if ((val3 <= thresh_val) && (val4 <= thresh_val)) Map2[dimX*dimY*k + i*dimY + j] = 1;                        
+			}
+		}
+	}
+	return 1;
+}
+
+float cleanMap(unsigned short *Map, int dimX, int dimY, int dimZ)
+{
+	int i, j, k, i1, j1, i2, j2, counter;
+#pragma omp parallel for shared(Map) private(i, j, k, i1, j1, i2, j2, counter)
+	for (i = 0; i<dimX; i++) {
+		for (j = 0; j<dimY; j++) {
+			for (k = 0; k<dimZ; k++) {
+
+				counter = 0;
+				for (i1 = -3; i1 <= 3; i1++) {
+					for (j1 = -3; j1 <= 3; j1++) {
+						i2 = i + i1;
+						j2 = j + j1;
+						if ((i2 >= 0) && (i2 < dimX) && (j2 >= 0) && (j2 < dimY)) {
+							if (Map[dimX*dimY*k + i2*dimY + j2] == 0) counter++;
+						}
+					}
+				}
+				if (counter < 24) Map[dimX*dimY*k + i*dimY + j] = 1;
+			}
+		}
+	}
+	return *Map;
+}
+
+/* Copy Image */
+float copyIm(float *A, float *U, int dimX, int dimY, int dimZ)
+{
+	int j;
+#pragma omp parallel for shared(A, U) private(j)
+	for (j = 0; j<dimX*dimY*dimZ; j++)  U[j] = A[j];
+	return *U;
+}
+/*********************3D *********************/
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/LLT_model_core.h b/main_func/regularizers_CPU/LLT_model_core.h
new file mode 100644
index 0000000..ef8803d
--- /dev/null
+++ b/main_func/regularizers_CPU/LLT_model_core.h
@@ -0,0 +1,64 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include <matrix.h>
+#include <math.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <stdio.h>
+#include "omp.h"
+
+#define EPS 0.01
+
+/* C-OMP implementation of Lysaker, Lundervold and Tai (LLT) model of higher order regularization penalty
+*
+* Input Parameters:
+* 1. U0 - origanal noise image/volume
+* 2. lambda - regularization parameter
+* 3. tau - time-step  for explicit scheme
+* 4. iter - iterations number
+* 5. epsil  - tolerance constant (to terminate earlier)
+* 6. switcher - default is 0, switch to (1) to restrictive smoothing in Z dimension (in test)
+*
+* Output:
+* Filtered/regularized image
+*
+* Example:
+* figure;
+* Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+* u0 = Im + .03*randn(size(Im)); % adding noise
+* [Den] = LLT_model(single(u0), 10, 0.1, 1);
+*
+*
+* to compile with OMP support: mex LLT_model.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
+* References: Lysaker, Lundervold and Tai (LLT) 2003, IEEE
+*
+* 28.11.16/Harwell
+*/
+/* 2D functions */
+float der2D(float *U, float *D1, float *D2, int dimX, int dimY, int dimZ);
+float div_upd2D(float *U0, float *U, float *D1, float *D2, int dimX, int dimY, int dimZ, float lambda, float tau);
+
+float der3D(float *U, float *D1, float *D2, float *D3, int dimX, int dimY, int dimZ);
+float div_upd3D(float *U0, float *U, float *D1, float *D2, float *D3, unsigned short *Map, int switcher, int dimX, int dimY, int dimZ, float lambda, float tau);
+
+float calcMap(float *U, unsigned short *Map, int dimX, int dimY, int dimZ);
+float cleanMap(unsigned short *Map, int dimX, int dimY, int dimZ);
+
+float copyIm(float *A, float *U, int dimX, int dimY, int dimZ);
diff --git a/main_func/regularizers_CPU/PatchBased_Regul.c b/main_func/regularizers_CPU/PatchBased_Regul.c
index 1ed29d4..24ee210 100644
--- a/main_func/regularizers_CPU/PatchBased_Regul.c
+++ b/main_func/regularizers_CPU/PatchBased_Regul.c
@@ -1,12 +1,25 @@
-#define _USE_MATH_DEFINES

+/*

+This work is part of the Core Imaging Library developed by

+Visual Analytics and Imaging System Group of the Science Technology

+Facilities Council, STFC

+

+Copyright 2017 Daniil Kazanteev

+Copyright 2017 Srikanth Nagella, Edoardo Pasca

+

+Licensed under the Apache License, Version 2.0 (the "License");

+you may not use this file except in compliance with the License.

+You may obtain a copy of the License at

+http://www.apache.org/licenses/LICENSE-2.0

+Unless required by applicable law or agreed to in writing, software

+distributed under the License is distributed on an "AS IS" BASIS,

+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

+See the License for the specific language governing permissions and

+limitations under the License.

+*/

 

 #include "mex.h"

-#include <matrix.h>

-#include <math.h>

-#include <stdlib.h>

-#include <memory.h>

-#include <stdio.h>

-#include "omp.h"

+#include "PatchBased_Regul_core.h"

+

 

 /* C-OMP implementation of  patch-based (PB) regularization (2D and 3D cases). 

  * This method finds self-similar patches in data and performs one fixed point iteration to mimimize the PB penalty function

@@ -41,9 +54,6 @@
  * Harwell, UK

  */

 

-float pad_crop(float *A, float *Ap, int OldSizeX, int OldSizeY,  int OldSizeZ, int NewSizeX, int NewSizeY, int NewSizeZ, int padXY, int switchpad_crop);

-float PB_FUNC2D(float *A, float *B, int dimX, int dimY, int padXY, int SearchW, int SimilW, float h, float lambda);

-float PB_FUNC3D(float *A, float *B, int dimX, int dimY, int dimZ, int padXY, int SearchW, int SimilW, float h, float lambda);      

 

 void mexFunction(

         int nlhs, mxArray *plhs[],

@@ -125,171 +135,4 @@ void mexFunction(
         switchpad_crop = 1; /*cropping*/

         pad_crop(Bp, B, M, N, Z, newsizeY, newsizeX, newsizeZ, padXY, switchpad_crop);

     } /*end else ndims*/ 

-}    

-    

-/*2D version*/

-float PB_FUNC2D(float *A, float *B, int dimX, int dimY, int padXY, int SearchW, int SimilW, float h, float lambda)

-{

-    int i, j, i_n, j_n, i_m, j_m, i_p, j_p, i_l, j_l, i1, j1, i2, j2, i3, j3, i5,j5, count, SimilW_full;

-    float *Eucl_Vec, h2, denh2, normsum, Weight, Weight_norm, value, denom, WeightGlob, t1;

-                 

-    /*SearchW_full = 2*SearchW + 1; */ /* the full searching window  size */

-    SimilW_full = 2*SimilW + 1;   /* the full similarity window  size */

-    h2 = h*h;

-    denh2 = 1/(2*h2);   

-    

-     /*Gaussian kernel */

-    Eucl_Vec = (float*) calloc (SimilW_full*SimilW_full,sizeof(float));

-    count = 0;

-    for(i_n=-SimilW; i_n<=SimilW; i_n++) {

-        for(j_n=-SimilW; j_n<=SimilW; j_n++) {

-            t1 = pow(((float)i_n), 2) + pow(((float)j_n), 2);

-            Eucl_Vec[count] = exp(-(t1)/(2*SimilW*SimilW));

-            count = count + 1;                       

-        }} /*main neighb loop */   

-    

-    /*The NLM code starts here*/         

-    /* setting OMP here */

-    #pragma omp parallel for shared (A, B, dimX, dimY, Eucl_Vec, lambda, denh2) private(denom, i, j, WeightGlob, count,  i1, j1, i2, j2, i3, j3, i5, j5, Weight_norm, normsum, i_m, j_m, i_n, j_n, i_l, j_l, i_p, j_p, Weight,  value)

-    

-    for(i=0; i<dimX; i++) {

-        for(j=0; j<dimY; j++) {

-             if (((i >= padXY) && (i < dimX-padXY)) &&  ((j >= padXY) && (j < dimY-padXY))) {

-          

-                /* Massive Search window loop */

-                Weight_norm = 0; value = 0.0;

-                for(i_m=-SearchW; i_m<=SearchW; i_m++) {

-                    for(j_m=-SearchW; j_m<=SearchW; j_m++) {

-                       /*checking boundaries*/

-                        i1 = i+i_m; j1 = j+j_m;

-                        

-                        WeightGlob = 0.0;

-                        /* if inside the searching window */                        

-                         for(i_l=-SimilW; i_l<=SimilW; i_l++) {

-                             for(j_l=-SimilW; j_l<=SimilW; j_l++) {

-                                 i2 = i1+i_l; j2 = j1+j_l;

-                                 

-                                 i3 = i+i_l; j3 = j+j_l;   /*coordinates of the inner patch loop */

-                                

-                                 count = 0; normsum = 0.0;

-                                 for(i_p=-SimilW; i_p<=SimilW; i_p++) {

-                                     for(j_p=-SimilW; j_p<=SimilW; j_p++) {

-                                         i5 = i2 + i_p; j5 = j2 + j_p;

-                                         normsum = normsum + Eucl_Vec[count]*pow(A[(i3+i_p)*dimY+(j3+j_p)]-A[i5*dimY+j5], 2);        

-                                         count = count + 1;

-                                     }}

-                                  if (normsum != 0) Weight = (exp(-normsum*denh2)); 

-                                  else Weight = 0.0;

-                                 WeightGlob += Weight;

-                             }}                      

-      

-                         value += A[i1*dimY+j1]*WeightGlob;

-                         Weight_norm += WeightGlob;           

-                    }}      /*search window loop end*/

-                

-                /* the final loop to average all values in searching window with weights */

-                denom = 1 + lambda*Weight_norm;

-                B[i*dimY+j] = (A[i*dimY+j] + lambda*value)/denom;         

-             }

-        }}   /*main loop*/     

-    return (*B);

-    free(Eucl_Vec);    

-}

- 

-/*3D version*/ 

- float PB_FUNC3D(float *A, float *B, int dimX, int dimY, int dimZ, int padXY, int SearchW, int SimilW, float h, float lambda)       

- {

-    int SimilW_full, count, i, j, k,  i_n, j_n, k_n, i_m, j_m, k_m, i_p, j_p, k_p, i_l, j_l, k_l, i1, j1, k1, i2, j2, k2, i3, j3, k3, i5, j5, k5;

-    float *Eucl_Vec, h2, denh2, normsum, Weight, Weight_norm, value, denom, WeightGlob;

-        

-    /*SearchW_full = 2*SearchW + 1; */ /* the full searching window  size */

-    SimilW_full = 2*SimilW + 1;   /* the full similarity window  size */

-    h2 = h*h;

-    denh2 = 1/(2*h2);

-    

-    /*Gaussian kernel */

-    Eucl_Vec = (float*) calloc (SimilW_full*SimilW_full*SimilW_full,sizeof(float));

-    count = 0;

-    for(i_n=-SimilW; i_n<=SimilW; i_n++) {

-        for(j_n=-SimilW; j_n<=SimilW; j_n++) {

-            for(k_n=-SimilW; k_n<=SimilW; k_n++) {

-                Eucl_Vec[count] = exp(-(pow((float)i_n, 2) + pow((float)j_n, 2) + pow((float)k_n, 2))/(2*SimilW*SimilW*SimilW));

-                count = count + 1;

-            }}} /*main neighb loop */

-    

-    /*The NLM code starts here*/         

-    /* setting OMP here */

-    #pragma omp parallel for shared (A, B, dimX, dimY, dimZ, Eucl_Vec, lambda, denh2) private(denom, i, j, k, WeightGlob,count,  i1, j1, k1, i2, j2, k2, i3, j3, k3, i5, j5, k5, Weight_norm, normsum, i_m, j_m, k_m, i_n, j_n, k_n, i_l, j_l, k_l, i_p, j_p, k_p, Weight, value)    

-    for(i=0; i<dimX; i++) {

-        for(j=0; j<dimY; j++) {

-            for(k=0; k<dimZ; k++) {

-            if (((i >= padXY) && (i < dimX-padXY)) &&  ((j >= padXY) && (j < dimY-padXY)) &&  ((k >= padXY) && (k < dimZ-padXY))) {

-            /* take all elements around the pixel of interest */                             

-               /* Massive Search window loop */

-                Weight_norm = 0;  value = 0.0;

-                for(i_m=-SearchW; i_m<=SearchW; i_m++) {

-                    for(j_m=-SearchW; j_m<=SearchW; j_m++) {

-                        for(k_m=-SearchW; k_m<=SearchW; k_m++) {

-                         /*checking boundaries*/

-                        i1 = i+i_m; j1 = j+j_m; k1 = k+k_m;

-                        

-                        WeightGlob = 0.0;

-                        /* if inside the searching window */                        

-                         for(i_l=-SimilW; i_l<=SimilW; i_l++) {

-                             for(j_l=-SimilW; j_l<=SimilW; j_l++) {

-                                 for(k_l=-SimilW; k_l<=SimilW; k_l++) {                                 

-                                 i2 = i1+i_l; j2 = j1+j_l; k2 = k1+k_l;

-                                 

-                                 i3 = i+i_l; j3 = j+j_l; k3 = k+k_l;   /*coordinates of the inner patch loop */

-                                

-                                 count = 0; normsum = 0.0;

-                                 for(i_p=-SimilW; i_p<=SimilW; i_p++) {

-                                     for(j_p=-SimilW; j_p<=SimilW; j_p++) {

-                                         for(k_p=-SimilW; k_p<=SimilW; k_p++) {

-                                         i5 = i2 + i_p; j5 = j2 + j_p; k5 = k2 + k_p;

-                                         normsum = normsum + Eucl_Vec[count]*pow(A[(dimX*dimY)*(k3+k_p)+(i3+i_p)*dimY+(j3+j_p)]-A[(dimX*dimY)*k5 + i5*dimY+j5], 2);        

-                                         count = count + 1;

-                                     }}}

-                                  if (normsum != 0) Weight = (exp(-normsum*denh2)); 

-                                  else Weight = 0.0;

-                                 WeightGlob += Weight;

-                             }}}                                                 

-                         value += A[(dimX*dimY)*k1 + i1*dimY+j1]*WeightGlob;

-                         Weight_norm += WeightGlob;

-             

-                    }}}      /*search window loop end*/

-                

-                /* the final loop to average all values in searching window with weights */

-                denom = 1 + lambda*Weight_norm;               

-                B[(dimX*dimY)*k + i*dimY+j] = (A[(dimX*dimY)*k + i*dimY+j] + lambda*value)/denom;      

-            }            

-        }}}   /*main loop*/              

-       free(Eucl_Vec);        

-       return *B;

-}

-

-float pad_crop(float *A, float *Ap, int OldSizeX, int OldSizeY, int OldSizeZ, int NewSizeX, int NewSizeY, int NewSizeZ, int padXY, int switchpad_crop)

-{

-    /* padding-cropping function */

-    int i,j,k;    

-    if (NewSizeZ > 1) {    

-           for (i=0; i < NewSizeX; i++) {

-            for (j=0; j < NewSizeY; j++) {

-              for (k=0; k < NewSizeZ; k++) {

-                if (((i >= padXY) && (i < NewSizeX-padXY)) &&  ((j >= padXY) && (j < NewSizeY-padXY)) &&  ((k >= padXY) && (k < NewSizeZ-padXY))) {

-                    if (switchpad_crop == 0)  Ap[NewSizeX*NewSizeY*k + i*NewSizeY+j] = A[OldSizeX*OldSizeY*(k - padXY) + (i-padXY)*(OldSizeY)+(j-padXY)];

-                    else  Ap[OldSizeX*OldSizeY*(k - padXY) + (i-padXY)*(OldSizeY)+(j-padXY)] = A[NewSizeX*NewSizeY*k + i*NewSizeY+j];

-                }

-            }}}   

-    }

-    else {

-        for (i=0; i < NewSizeX; i++) {

-            for (j=0; j < NewSizeY; j++) {

-                if (((i >= padXY) && (i < NewSizeX-padXY)) &&  ((j >= padXY) && (j < NewSizeY-padXY))) {

-                    if (switchpad_crop == 0)  Ap[i*NewSizeY+j] = A[(i-padXY)*(OldSizeY)+(j-padXY)];

-                    else  Ap[(i-padXY)*(OldSizeY)+(j-padXY)] = A[i*NewSizeY+j];

-                }

-            }}

-    }

-    return *Ap;

-}
\ No newline at end of file
+}    
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/PatchBased_Regul_core.c b/main_func/regularizers_CPU/PatchBased_Regul_core.c
new file mode 100644
index 0000000..6f0a48d
--- /dev/null
+++ b/main_func/regularizers_CPU/PatchBased_Regul_core.c
@@ -0,0 +1,220 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include "PatchBased_Regul_core.h"
+
+/* C-OMP implementation of  patch-based (PB) regularization (2D and 3D cases).
+* This method finds self-similar patches in data and performs one fixed point iteration to mimimize the PB penalty function
+*
+* References: 1. Yang Z. & Jacob M. "Nonlocal Regularization of Inverse Problems"
+*             2. Kazantsev D. et al. "4D-CT reconstruction with unified spatial-temporal patch-based regularization"
+*
+* Input Parameters (mandatory):
+* 1. Image (2D or 3D)
+* 2. ratio of the searching window (e.g. 3 = (2*3+1) = 7 pixels window)
+* 3. ratio of the similarity window (e.g. 1 = (2*1+1) = 3 pixels window)
+* 4. h - parameter for the PB penalty function
+* 5. lambda - regularization parameter
+
+* Output:
+* 1. regularized (denoised) Image (N x N)/volume (N x N x N)
+*
+* Quick 2D denoising example in Matlab:
+Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+u0 = Im + .03*randn(size(Im)); u0(u0<0) = 0; % adding noise
+ImDen = PB_Regul_CPU(single(u0), 3, 1, 0.08, 0.05);
+*
+* Please see more tests in a file:
+TestTemporalSmoothing.m
+
+*
+* Matlab + C/mex compilers needed
+* to compile with OMP support: mex PB_Regul_CPU.c CFLAGS="\$CFLAGS -fopenmp -Wall" LDFLAGS="\$LDFLAGS -fopenmp"
+*
+* D. Kazantsev *
+* 02/07/2014
+* Harwell, UK
+*/
+
+/*2D version*/
+float PB_FUNC2D(float *A, float *B, int dimX, int dimY, int padXY, int SearchW, int SimilW, float h, float lambda)
+{
+    int i, j, i_n, j_n, i_m, j_m, i_p, j_p, i_l, j_l, i1, j1, i2, j2, i3, j3, i5,j5, count, SimilW_full;
+    float *Eucl_Vec, h2, denh2, normsum, Weight, Weight_norm, value, denom, WeightGlob, t1;
+                 
+    /*SearchW_full = 2*SearchW + 1; */ /* the full searching window  size */
+    SimilW_full = 2*SimilW + 1;   /* the full similarity window  size */
+    h2 = h*h;
+    denh2 = 1/(2*h2);   
+    
+     /*Gaussian kernel */
+    Eucl_Vec = (float*) calloc (SimilW_full*SimilW_full,sizeof(float));
+    count = 0;
+    for(i_n=-SimilW; i_n<=SimilW; i_n++) {
+        for(j_n=-SimilW; j_n<=SimilW; j_n++) {
+            t1 = pow(((float)i_n), 2) + pow(((float)j_n), 2);
+            Eucl_Vec[count] = exp(-(t1)/(2*SimilW*SimilW));
+            count = count + 1;                       
+        }} /*main neighb loop */   
+    
+    /*The NLM code starts here*/         
+    /* setting OMP here */
+    #pragma omp parallel for shared (A, B, dimX, dimY, Eucl_Vec, lambda, denh2) private(denom, i, j, WeightGlob, count,  i1, j1, i2, j2, i3, j3, i5, j5, Weight_norm, normsum, i_m, j_m, i_n, j_n, i_l, j_l, i_p, j_p, Weight,  value)
+    
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+             if (((i >= padXY) && (i < dimX-padXY)) &&  ((j >= padXY) && (j < dimY-padXY))) {
+          
+                /* Massive Search window loop */
+                Weight_norm = 0; value = 0.0;
+                for(i_m=-SearchW; i_m<=SearchW; i_m++) {
+                    for(j_m=-SearchW; j_m<=SearchW; j_m++) {
+                       /*checking boundaries*/
+                        i1 = i+i_m; j1 = j+j_m;
+                        
+                        WeightGlob = 0.0;
+                        /* if inside the searching window */                        
+                         for(i_l=-SimilW; i_l<=SimilW; i_l++) {
+                             for(j_l=-SimilW; j_l<=SimilW; j_l++) {
+                                 i2 = i1+i_l; j2 = j1+j_l;
+                                 
+                                 i3 = i+i_l; j3 = j+j_l;   /*coordinates of the inner patch loop */
+                                
+                                 count = 0; normsum = 0.0;
+                                 for(i_p=-SimilW; i_p<=SimilW; i_p++) {
+                                     for(j_p=-SimilW; j_p<=SimilW; j_p++) {
+                                         i5 = i2 + i_p; j5 = j2 + j_p;
+                                         normsum = normsum + Eucl_Vec[count]*pow(A[(i3+i_p)*dimY+(j3+j_p)]-A[i5*dimY+j5], 2);        
+                                         count = count + 1;
+                                     }}
+                                  if (normsum != 0) Weight = (exp(-normsum*denh2)); 
+                                  else Weight = 0.0;
+                                 WeightGlob += Weight;
+                             }}                      
+      
+                         value += A[i1*dimY+j1]*WeightGlob;
+                         Weight_norm += WeightGlob;           
+                    }}      /*search window loop end*/
+                
+                /* the final loop to average all values in searching window with weights */
+                denom = 1 + lambda*Weight_norm;
+                B[i*dimY+j] = (A[i*dimY+j] + lambda*value)/denom;         
+             }
+        }}   /*main loop*/     
+    return (*B);
+    free(Eucl_Vec);    
+}
+ 
+/*3D version*/ 
+ float PB_FUNC3D(float *A, float *B, int dimX, int dimY, int dimZ, int padXY, int SearchW, int SimilW, float h, float lambda)       
+ {
+    int SimilW_full, count, i, j, k,  i_n, j_n, k_n, i_m, j_m, k_m, i_p, j_p, k_p, i_l, j_l, k_l, i1, j1, k1, i2, j2, k2, i3, j3, k3, i5, j5, k5;
+    float *Eucl_Vec, h2, denh2, normsum, Weight, Weight_norm, value, denom, WeightGlob;
+        
+    /*SearchW_full = 2*SearchW + 1; */ /* the full searching window  size */
+    SimilW_full = 2*SimilW + 1;   /* the full similarity window  size */
+    h2 = h*h;
+    denh2 = 1/(2*h2);
+    
+    /*Gaussian kernel */
+    Eucl_Vec = (float*) calloc (SimilW_full*SimilW_full*SimilW_full,sizeof(float));
+    count = 0;
+    for(i_n=-SimilW; i_n<=SimilW; i_n++) {
+        for(j_n=-SimilW; j_n<=SimilW; j_n++) {
+            for(k_n=-SimilW; k_n<=SimilW; k_n++) {
+                Eucl_Vec[count] = exp(-(pow((float)i_n, 2) + pow((float)j_n, 2) + pow((float)k_n, 2))/(2*SimilW*SimilW*SimilW));
+                count = count + 1;
+            }}} /*main neighb loop */
+    
+    /*The NLM code starts here*/         
+    /* setting OMP here */
+    #pragma omp parallel for shared (A, B, dimX, dimY, dimZ, Eucl_Vec, lambda, denh2) private(denom, i, j, k, WeightGlob,count,  i1, j1, k1, i2, j2, k2, i3, j3, k3, i5, j5, k5, Weight_norm, normsum, i_m, j_m, k_m, i_n, j_n, k_n, i_l, j_l, k_l, i_p, j_p, k_p, Weight, value)    
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            for(k=0; k<dimZ; k++) {
+            if (((i >= padXY) && (i < dimX-padXY)) &&  ((j >= padXY) && (j < dimY-padXY)) &&  ((k >= padXY) && (k < dimZ-padXY))) {
+            /* take all elements around the pixel of interest */                             
+               /* Massive Search window loop */
+                Weight_norm = 0;  value = 0.0;
+                for(i_m=-SearchW; i_m<=SearchW; i_m++) {
+                    for(j_m=-SearchW; j_m<=SearchW; j_m++) {
+                        for(k_m=-SearchW; k_m<=SearchW; k_m++) {
+                         /*checking boundaries*/
+                        i1 = i+i_m; j1 = j+j_m; k1 = k+k_m;
+                        
+                        WeightGlob = 0.0;
+                        /* if inside the searching window */                        
+                         for(i_l=-SimilW; i_l<=SimilW; i_l++) {
+                             for(j_l=-SimilW; j_l<=SimilW; j_l++) {
+                                 for(k_l=-SimilW; k_l<=SimilW; k_l++) {                                 
+                                 i2 = i1+i_l; j2 = j1+j_l; k2 = k1+k_l;
+                                 
+                                 i3 = i+i_l; j3 = j+j_l; k3 = k+k_l;   /*coordinates of the inner patch loop */
+                                
+                                 count = 0; normsum = 0.0;
+                                 for(i_p=-SimilW; i_p<=SimilW; i_p++) {
+                                     for(j_p=-SimilW; j_p<=SimilW; j_p++) {
+                                         for(k_p=-SimilW; k_p<=SimilW; k_p++) {
+                                         i5 = i2 + i_p; j5 = j2 + j_p; k5 = k2 + k_p;
+                                         normsum = normsum + Eucl_Vec[count]*pow(A[(dimX*dimY)*(k3+k_p)+(i3+i_p)*dimY+(j3+j_p)]-A[(dimX*dimY)*k5 + i5*dimY+j5], 2);        
+                                         count = count + 1;
+                                     }}}
+                                  if (normsum != 0) Weight = (exp(-normsum*denh2)); 
+                                  else Weight = 0.0;
+                                 WeightGlob += Weight;
+                             }}}                                                 
+                         value += A[(dimX*dimY)*k1 + i1*dimY+j1]*WeightGlob;
+                         Weight_norm += WeightGlob;
+             
+                    }}}      /*search window loop end*/
+                
+                /* the final loop to average all values in searching window with weights */
+                denom = 1 + lambda*Weight_norm;               
+                B[(dimX*dimY)*k + i*dimY+j] = (A[(dimX*dimY)*k + i*dimY+j] + lambda*value)/denom;      
+            }            
+        }}}   /*main loop*/              
+       free(Eucl_Vec);        
+       return *B;
+}
+
+float pad_crop(float *A, float *Ap, int OldSizeX, int OldSizeY, int OldSizeZ, int NewSizeX, int NewSizeY, int NewSizeZ, int padXY, int switchpad_crop)
+{
+    /* padding-cropping function */
+    int i,j,k;    
+    if (NewSizeZ > 1) {    
+           for (i=0; i < NewSizeX; i++) {
+            for (j=0; j < NewSizeY; j++) {
+              for (k=0; k < NewSizeZ; k++) {
+                if (((i >= padXY) && (i < NewSizeX-padXY)) &&  ((j >= padXY) && (j < NewSizeY-padXY)) &&  ((k >= padXY) && (k < NewSizeZ-padXY))) {
+                    if (switchpad_crop == 0)  Ap[NewSizeX*NewSizeY*k + i*NewSizeY+j] = A[OldSizeX*OldSizeY*(k - padXY) + (i-padXY)*(OldSizeY)+(j-padXY)];
+                    else  Ap[OldSizeX*OldSizeY*(k - padXY) + (i-padXY)*(OldSizeY)+(j-padXY)] = A[NewSizeX*NewSizeY*k + i*NewSizeY+j];
+                }
+            }}}   
+    }
+    else {
+        for (i=0; i < NewSizeX; i++) {
+            for (j=0; j < NewSizeY; j++) {
+                if (((i >= padXY) && (i < NewSizeX-padXY)) &&  ((j >= padXY) && (j < NewSizeY-padXY))) {
+                    if (switchpad_crop == 0)  Ap[i*NewSizeY+j] = A[(i-padXY)*(OldSizeY)+(j-padXY)];
+                    else  Ap[(i-padXY)*(OldSizeY)+(j-padXY)] = A[i*NewSizeY+j];
+                }
+            }}
+    }
+    return *Ap;
+}
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/PatchBased_Regul_core.h b/main_func/regularizers_CPU/PatchBased_Regul_core.h
new file mode 100644
index 0000000..b83cf10
--- /dev/null
+++ b/main_func/regularizers_CPU/PatchBased_Regul_core.h
@@ -0,0 +1,64 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#define _USE_MATH_DEFINES
+
+#include <matrix.h>
+#include <math.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <stdio.h>
+#include "omp.h"
+
+/* C-OMP implementation of  patch-based (PB) regularization (2D and 3D cases).
+* This method finds self-similar patches in data and performs one fixed point iteration to mimimize the PB penalty function
+*
+* References: 1. Yang Z. & Jacob M. "Nonlocal Regularization of Inverse Problems"
+*             2. Kazantsev D. et al. "4D-CT reconstruction with unified spatial-temporal patch-based regularization"
+*
+* Input Parameters (mandatory):
+* 1. Image (2D or 3D)
+* 2. ratio of the searching window (e.g. 3 = (2*3+1) = 7 pixels window)
+* 3. ratio of the similarity window (e.g. 1 = (2*1+1) = 3 pixels window)
+* 4. h - parameter for the PB penalty function
+* 5. lambda - regularization parameter
+
+* Output:
+* 1. regularized (denoised) Image (N x N)/volume (N x N x N)
+*
+* Quick 2D denoising example in Matlab:
+Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+u0 = Im + .03*randn(size(Im)); u0(u0<0) = 0; % adding noise
+ImDen = PB_Regul_CPU(single(u0), 3, 1, 0.08, 0.05);
+*
+* Please see more tests in a file:
+TestTemporalSmoothing.m
+
+*
+* Matlab + C/mex compilers needed
+* to compile with OMP support: mex PB_Regul_CPU.c CFLAGS="\$CFLAGS -fopenmp -Wall" LDFLAGS="\$LDFLAGS -fopenmp"
+*
+* D. Kazantsev *
+* 02/07/2014
+* Harwell, UK
+*/
+
+float pad_crop(float *A, float *Ap, int OldSizeX, int OldSizeY, int OldSizeZ, int NewSizeX, int NewSizeY, int NewSizeZ, int padXY, int switchpad_crop);
+float PB_FUNC2D(float *A, float *B, int dimX, int dimY, int padXY, int SearchW, int SimilW, float h, float lambda);
+float PB_FUNC3D(float *A, float *B, int dimX, int dimY, int dimZ, int padXY, int SearchW, int SimilW, float h, float lambda);
diff --git a/main_func/regularizers_CPU/SplitBregman_TV.c b/main_func/regularizers_CPU/SplitBregman_TV.c
index f143aa6..0dc638d 100644
--- a/main_func/regularizers_CPU/SplitBregman_TV.c
+++ b/main_func/regularizers_CPU/SplitBregman_TV.c
@@ -1,10 +1,24 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
 #include "mex.h"
-#include <matrix.h>
-#include <math.h>
-#include <stdlib.h>
-#include <memory.h>
-#include <stdio.h>
-#include "omp.h"
+#include "SplitBregman_TV_core.h"
 
 /* C-OMP implementation of Split Bregman - TV denoising-regularization model (2D/3D)
  *
@@ -30,16 +44,6 @@
  * D. Kazantsev, 2016*
  */
 
-float copyIm(float *A, float *B, int dimX, int dimY, int dimZ);
-float gauss_seidel2D(float *U,  float *A, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda, float mu);
-float updDxDy_shrinkAniso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda);
-float updDxDy_shrinkIso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda);
-float updBxBy2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY);
-
-float gauss_seidel3D(float *U, float *A, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda, float mu);
-float updDxDyDz_shrinkAniso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda);
-float updDxDyDz_shrinkIso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda);
-float updBxByBz3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ);
 
 void mexFunction(
         int nlhs, mxArray *plhs[],
@@ -171,229 +175,4 @@ void mexFunction(
         }
         printf("SB iterations stopped at iteration: %i\n", ll);
     }
-}
-
-/* 2D-case related Functions */
-/*****************************************************************/
-float gauss_seidel2D(float *U, float *A, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda, float mu)
-{
-    float sum, normConst;
-    int i,j,i1,i2,j1,j2;
-    normConst = 1.0f/(mu + 4.0f*lambda);
-    
-#pragma omp parallel for shared(U) private(i,j,i1,i2,j1,j2,sum)
-    for(i=0; i<dimX; i++) {
-        /* symmetric boundary conditions (Neuman) */
-        i1 = i+1; if (i1 == dimX) i1 = i-1;
-        i2 = i-1; if (i2 < 0) i2 = i+1;
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            j1 = j+1; if (j1 == dimY) j1 = j-1;
-            j2 = j-1; if (j2 < 0) j2 = j+1;
-            
-            sum = Dx[(i2)*dimY + (j)] - Dx[(i)*dimY + (j)] + Dy[(i)*dimY + (j2)] - Dy[(i)*dimY + (j)] - Bx[(i2)*dimY + (j)] + Bx[(i)*dimY + (j)] - By[(i)*dimY + (j2)] + By[(i)*dimY + (j)];
-            sum += (U[(i1)*dimY + (j)] + U[(i2)*dimY + (j)] + U[(i)*dimY + (j1)] + U[(i)*dimY + (j2)]);
-            sum *= lambda;
-            sum += mu*A[(i)*dimY + (j)];
-            U[(i)*dimY + (j)] = normConst*sum;
-        }}
-    return *U;
-}
-
-float updDxDy_shrinkAniso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda)
-{
-    int i,j,i1,j1;
-    float val1, val11, val2, val22, denom_lam;
-    denom_lam = 1.0f/lambda;
-#pragma omp parallel for shared(U,denom_lam) private(i,j,i1,j1,val1,val11,val2,val22)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            i1 = i+1; if (i1 == dimX) i1 = i-1;
-            j1 = j+1; if (j1 == dimY) j1 = j-1;
-            
-            val1 = (U[(i1)*dimY + (j)] - U[(i)*dimY + (j)]) + Bx[(i)*dimY + (j)];
-            val2 = (U[(i)*dimY + (j1)] - U[(i)*dimY + (j)]) + By[(i)*dimY + (j)];
-            
-            val11 = fabs(val1) - denom_lam; if (val11 < 0) val11 = 0;
-            val22 = fabs(val2) - denom_lam; if (val22 < 0) val22 = 0;
-            
-            if (val1 !=0) Dx[(i)*dimY + (j)] = (val1/fabs(val1))*val11; else Dx[(i)*dimY + (j)] = 0;
-            if (val2 !=0) Dy[(i)*dimY + (j)] = (val2/fabs(val2))*val22; else Dy[(i)*dimY + (j)] = 0;
-            
-        }}
-    return 1;
-}
-float updDxDy_shrinkIso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda)
-{
-    int i,j,i1,j1;
-    float val1, val11, val2, denom, denom_lam;
-    denom_lam = 1.0f/lambda;
-    
-#pragma omp parallel for shared(U,denom_lam) private(i,j,i1,j1,val1,val11,val2,denom)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            i1 = i+1; if (i1 == dimX) i1 = i-1;
-            j1 = j+1; if (j1 == dimY) j1 = j-1;
-            
-            val1 = (U[(i1)*dimY + (j)] - U[(i)*dimY + (j)]) + Bx[(i)*dimY + (j)];
-            val2 = (U[(i)*dimY + (j1)] - U[(i)*dimY + (j)]) + By[(i)*dimY + (j)];
-            
-            denom = sqrt(val1*val1 + val2*val2);
-            
-            val11 = (denom - denom_lam); if (val11 < 0) val11 = 0.0f;
-            
-            if (denom != 0.0f) {
-                Dx[(i)*dimY + (j)] = val11*(val1/denom);
-                Dy[(i)*dimY + (j)] = val11*(val2/denom);
-            }
-            else {
-                Dx[(i)*dimY + (j)] = 0;
-                Dy[(i)*dimY + (j)] = 0;
-            }
-        }}
-    return 1;
-}
-float updBxBy2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY)
-{
-    int i,j,i1,j1;
-#pragma omp parallel for shared(U) private(i,j,i1,j1)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            i1 = i+1; if (i1 == dimX) i1 = i-1;
-            j1 = j+1; if (j1 == dimY) j1 = j-1;
-            
-            Bx[(i)*dimY + (j)] = Bx[(i)*dimY + (j)] + ((U[(i1)*dimY + (j)] - U[(i)*dimY + (j)]) - Dx[(i)*dimY + (j)]);
-            By[(i)*dimY + (j)] = By[(i)*dimY + (j)] + ((U[(i)*dimY + (j1)] - U[(i)*dimY + (j)]) - Dy[(i)*dimY + (j)]);
-        }}
-    return 1;
-}
-
-
-/* 3D-case related Functions */
-/*****************************************************************/
-float gauss_seidel3D(float *U, float *A, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda, float mu)
-{
-    float normConst, d_val, b_val, sum;
-    int i,j,i1,i2,j1,j2,k,k1,k2;
-    normConst = 1.0f/(mu + 6.0f*lambda);
-#pragma omp parallel for shared(U) private(i,j,i1,i2,j1,j2,k,k1,k2,d_val,b_val,sum)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                /* symmetric boundary conditions (Neuman) */
-                i1 = i+1; if (i1 == dimX) i1 = i-1;
-                i2 = i-1; if (i2 < 0) i2 = i+1;
-                j1 = j+1; if (j1 == dimY) j1 = j-1;
-                j2 = j-1; if (j2 < 0) j2 = j+1;
-                k1 = k+1; if (k1 == dimZ) k1 = k-1;
-                k2 = k-1; if (k2 < 0) k2 = k+1;
-                
-                d_val = Dx[(dimX*dimY)*k + (i2)*dimY + (j)] - Dx[(dimX*dimY)*k + (i)*dimY + (j)] + Dy[(dimX*dimY)*k + (i)*dimY + (j2)] - Dy[(dimX*dimY)*k + (i)*dimY + (j)] + Dz[(dimX*dimY)*k2 + (i)*dimY + (j)] - Dz[(dimX*dimY)*k + (i)*dimY + (j)];
-                b_val = -Bx[(dimX*dimY)*k + (i2)*dimY + (j)] + Bx[(dimX*dimY)*k + (i)*dimY + (j)] - By[(dimX*dimY)*k + (i)*dimY + (j2)] + By[(dimX*dimY)*k + (i)*dimY + (j)] - Bz[(dimX*dimY)*k2 + (i)*dimY + (j)] + Bz[(dimX*dimY)*k + (i)*dimY + (j)];
-                sum =  d_val + b_val;
-                sum += U[(dimX*dimY)*k + (i1)*dimY + (j)] + U[(dimX*dimY)*k + (i2)*dimY + (j)] + U[(dimX*dimY)*k + (i)*dimY + (j1)] + U[(dimX*dimY)*k + (i)*dimY + (j2)] + U[(dimX*dimY)*k1 + (i)*dimY + (j)] + U[(dimX*dimY)*k2 + (i)*dimY + (j)];
-                sum *= lambda;
-                sum += mu*A[(dimX*dimY)*k + (i)*dimY + (j)];
-                U[(dimX*dimY)*k + (i)*dimY + (j)] = normConst*sum;
-            }}}
-    return *U;
-}
-
-float updDxDyDz_shrinkAniso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda)
-{
-    int i,j,i1,j1,k,k1,index;
-    float val1, val11, val2, val22, val3, val33, denom_lam;
-    denom_lam = 1.0f/lambda;
-#pragma omp parallel for shared(U,denom_lam) private(index,i,j,i1,j1,k,k1,val1,val11,val2,val22,val3,val33)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                index = (dimX*dimY)*k + (i)*dimY + (j);
-                /* symmetric boundary conditions (Neuman) */
-                i1 = i+1; if (i1 == dimX) i1 = i-1;
-                j1 = j+1; if (j1 == dimY) j1 = j-1;
-                k1 = k+1; if (k1 == dimZ) k1 = k-1;
-                
-                val1 = (U[(dimX*dimY)*k + (i1)*dimY + (j)] - U[index]) + Bx[index];
-                val2 = (U[(dimX*dimY)*k + (i)*dimY + (j1)] - U[index]) + By[index];
-                val3 = (U[(dimX*dimY)*k1 + (i)*dimY + (j)] - U[index]) + Bz[index];
-                
-                val11 = fabs(val1) - denom_lam; if (val11 < 0) val11 = 0;
-                val22 = fabs(val2) - denom_lam; if (val22 < 0) val22 = 0;
-                val33 = fabs(val3) - denom_lam; if (val33 < 0) val33 = 0;
-                
-                if (val1 !=0) Dx[index] = (val1/fabs(val1))*val11; else Dx[index] = 0;
-                if (val2 !=0) Dy[index] = (val2/fabs(val2))*val22; else Dy[index] = 0;
-                if (val3 !=0) Dz[index] = (val3/fabs(val3))*val33; else Dz[index] = 0;
-                
-            }}}
-    return 1;
-}
-float updDxDyDz_shrinkIso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda)
-{
-    int i,j,i1,j1,k,k1,index;
-    float val1, val11, val2, val3, denom, denom_lam;
-    denom_lam = 1.0f/lambda;
-#pragma omp parallel for shared(U,denom_lam) private(index,denom,i,j,i1,j1,k,k1,val1,val11,val2,val3)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                index = (dimX*dimY)*k + (i)*dimY + (j);
-                /* symmetric boundary conditions (Neuman) */
-                i1 = i+1; if (i1 == dimX) i1 = i-1;
-                j1 = j+1; if (j1 == dimY) j1 = j-1;
-                k1 = k+1; if (k1 == dimZ) k1 = k-1;
-                
-                val1 = (U[(dimX*dimY)*k + (i1)*dimY + (j)] - U[index]) + Bx[index];
-                val2 = (U[(dimX*dimY)*k + (i)*dimY + (j1)] - U[index]) + By[index];
-                val3 = (U[(dimX*dimY)*k1 + (i)*dimY + (j)] - U[index]) + Bz[index];
-                
-                denom = sqrt(val1*val1 + val2*val2 + val3*val3);
-                
-                val11 = (denom - denom_lam); if (val11 < 0) val11 = 0.0f;
-                
-                if (denom != 0.0f) {
-                    Dx[index] = val11*(val1/denom);
-                    Dy[index] = val11*(val2/denom);
-                    Dz[index] = val11*(val3/denom);
-                }
-                else {
-                    Dx[index] = 0;
-                    Dy[index] = 0;
-                    Dz[index] = 0;
-                }               
-            }}}
-    return 1;
-}
-float updBxByBz3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ)
-{
-    int i,j,k,i1,j1,k1;
-#pragma omp parallel for shared(U) private(i,j,k,i1,j1,k1)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                /* symmetric boundary conditions (Neuman) */
-                i1 = i+1; if (i1 == dimX) i1 = i-1;
-                j1 = j+1; if (j1 == dimY) j1 = j-1;
-                k1 = k+1; if (k1 == dimZ) k1 = k-1;
-                
-                Bx[(dimX*dimY)*k + (i)*dimY + (j)] = Bx[(dimX*dimY)*k + (i)*dimY + (j)] + ((U[(dimX*dimY)*k + (i1)*dimY + (j)] - U[(dimX*dimY)*k + (i)*dimY + (j)]) - Dx[(dimX*dimY)*k + (i)*dimY + (j)]);
-                By[(dimX*dimY)*k + (i)*dimY + (j)] = By[(dimX*dimY)*k + (i)*dimY + (j)] + ((U[(dimX*dimY)*k + (i)*dimY + (j1)] - U[(dimX*dimY)*k + (i)*dimY + (j)]) - Dy[(dimX*dimY)*k + (i)*dimY + (j)]);
-                Bz[(dimX*dimY)*k + (i)*dimY + (j)] = Bz[(dimX*dimY)*k + (i)*dimY + (j)] + ((U[(dimX*dimY)*k1 + (i)*dimY + (j)] - U[(dimX*dimY)*k + (i)*dimY + (j)]) - Dz[(dimX*dimY)*k + (i)*dimY + (j)]);
-                
-            }}}
-    return 1;
-}
-/* General Functions */
-/*****************************************************************/
-/* Copy Image */
-float copyIm(float *A, float *B, int dimX, int dimY, int dimZ)
-{
-    int j;
-#pragma omp parallel for shared(A, B) private(j)
-    for(j=0; j<dimX*dimY*dimZ; j++)  B[j] = A[j];
-    return *B;
 }
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/SplitBregman_TV_core.c b/main_func/regularizers_CPU/SplitBregman_TV_core.c
new file mode 100644
index 0000000..26ad5b1
--- /dev/null
+++ b/main_func/regularizers_CPU/SplitBregman_TV_core.c
@@ -0,0 +1,270 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include "SplitBregman_TV_core.h"
+
+/* C-OMP implementation of Split Bregman - TV denoising-regularization model (2D/3D)
+*
+* Input Parameters:
+* 1. Noisy image/volume
+* 2. lambda - regularization parameter
+* 3. Number of iterations [OPTIONAL parameter]
+* 4. eplsilon - tolerance constant [OPTIONAL parameter]
+* 5. TV-type: 'iso' or 'l1' [OPTIONAL parameter]
+*
+* Output:
+* Filtered/regularized image
+*
+* Example:
+* figure;
+* Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+* u0 = Im + .05*randn(size(Im)); u0(u0 < 0) = 0;
+* u = SplitBregman_TV(single(u0), 10, 30, 1e-04);
+*
+* to compile with OMP support: mex SplitBregman_TV.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
+* References:
+* The Split Bregman Method for L1 Regularized Problems, by Tom Goldstein and Stanley Osher.
+* D. Kazantsev, 2016*
+*/
+
+
+/* 2D-case related Functions */
+/*****************************************************************/
+float gauss_seidel2D(float *U, float *A, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda, float mu)
+{
+    float sum, normConst;
+    int i,j,i1,i2,j1,j2;
+    normConst = 1.0f/(mu + 4.0f*lambda);
+    
+#pragma omp parallel for shared(U) private(i,j,i1,i2,j1,j2,sum)
+    for(i=0; i<dimX; i++) {
+        /* symmetric boundary conditions (Neuman) */
+        i1 = i+1; if (i1 == dimX) i1 = i-1;
+        i2 = i-1; if (i2 < 0) i2 = i+1;
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            j1 = j+1; if (j1 == dimY) j1 = j-1;
+            j2 = j-1; if (j2 < 0) j2 = j+1;
+            
+            sum = Dx[(i2)*dimY + (j)] - Dx[(i)*dimY + (j)] + Dy[(i)*dimY + (j2)] - Dy[(i)*dimY + (j)] - Bx[(i2)*dimY + (j)] + Bx[(i)*dimY + (j)] - By[(i)*dimY + (j2)] + By[(i)*dimY + (j)];
+            sum += (U[(i1)*dimY + (j)] + U[(i2)*dimY + (j)] + U[(i)*dimY + (j1)] + U[(i)*dimY + (j2)]);
+            sum *= lambda;
+            sum += mu*A[(i)*dimY + (j)];
+            U[(i)*dimY + (j)] = normConst*sum;
+        }}
+    return *U;
+}
+
+float updDxDy_shrinkAniso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda)
+{
+    int i,j,i1,j1;
+    float val1, val11, val2, val22, denom_lam;
+    denom_lam = 1.0f/lambda;
+#pragma omp parallel for shared(U,denom_lam) private(i,j,i1,j1,val1,val11,val2,val22)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            i1 = i+1; if (i1 == dimX) i1 = i-1;
+            j1 = j+1; if (j1 == dimY) j1 = j-1;
+            
+            val1 = (U[(i1)*dimY + (j)] - U[(i)*dimY + (j)]) + Bx[(i)*dimY + (j)];
+            val2 = (U[(i)*dimY + (j1)] - U[(i)*dimY + (j)]) + By[(i)*dimY + (j)];
+            
+            val11 = fabs(val1) - denom_lam; if (val11 < 0) val11 = 0;
+            val22 = fabs(val2) - denom_lam; if (val22 < 0) val22 = 0;
+            
+            if (val1 !=0) Dx[(i)*dimY + (j)] = (val1/fabs(val1))*val11; else Dx[(i)*dimY + (j)] = 0;
+            if (val2 !=0) Dy[(i)*dimY + (j)] = (val2/fabs(val2))*val22; else Dy[(i)*dimY + (j)] = 0;
+            
+        }}
+    return 1;
+}
+float updDxDy_shrinkIso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda)
+{
+    int i,j,i1,j1;
+    float val1, val11, val2, denom, denom_lam;
+    denom_lam = 1.0f/lambda;
+    
+#pragma omp parallel for shared(U,denom_lam) private(i,j,i1,j1,val1,val11,val2,denom)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            i1 = i+1; if (i1 == dimX) i1 = i-1;
+            j1 = j+1; if (j1 == dimY) j1 = j-1;
+            
+            val1 = (U[(i1)*dimY + (j)] - U[(i)*dimY + (j)]) + Bx[(i)*dimY + (j)];
+            val2 = (U[(i)*dimY + (j1)] - U[(i)*dimY + (j)]) + By[(i)*dimY + (j)];
+            
+            denom = sqrt(val1*val1 + val2*val2);
+            
+            val11 = (denom - denom_lam); if (val11 < 0) val11 = 0.0f;
+            
+            if (denom != 0.0f) {
+                Dx[(i)*dimY + (j)] = val11*(val1/denom);
+                Dy[(i)*dimY + (j)] = val11*(val2/denom);
+            }
+            else {
+                Dx[(i)*dimY + (j)] = 0;
+                Dy[(i)*dimY + (j)] = 0;
+            }
+        }}
+    return 1;
+}
+float updBxBy2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY)
+{
+    int i,j,i1,j1;
+#pragma omp parallel for shared(U) private(i,j,i1,j1)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            i1 = i+1; if (i1 == dimX) i1 = i-1;
+            j1 = j+1; if (j1 == dimY) j1 = j-1;
+            
+            Bx[(i)*dimY + (j)] = Bx[(i)*dimY + (j)] + ((U[(i1)*dimY + (j)] - U[(i)*dimY + (j)]) - Dx[(i)*dimY + (j)]);
+            By[(i)*dimY + (j)] = By[(i)*dimY + (j)] + ((U[(i)*dimY + (j1)] - U[(i)*dimY + (j)]) - Dy[(i)*dimY + (j)]);
+        }}
+    return 1;
+}
+
+
+/* 3D-case related Functions */
+/*****************************************************************/
+float gauss_seidel3D(float *U, float *A, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda, float mu)
+{
+    float normConst, d_val, b_val, sum;
+    int i,j,i1,i2,j1,j2,k,k1,k2;
+    normConst = 1.0f/(mu + 6.0f*lambda);
+#pragma omp parallel for shared(U) private(i,j,i1,i2,j1,j2,k,k1,k2,d_val,b_val,sum)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            for(k=0; k<dimZ; k++) {
+                /* symmetric boundary conditions (Neuman) */
+                i1 = i+1; if (i1 == dimX) i1 = i-1;
+                i2 = i-1; if (i2 < 0) i2 = i+1;
+                j1 = j+1; if (j1 == dimY) j1 = j-1;
+                j2 = j-1; if (j2 < 0) j2 = j+1;
+                k1 = k+1; if (k1 == dimZ) k1 = k-1;
+                k2 = k-1; if (k2 < 0) k2 = k+1;
+                
+                d_val = Dx[(dimX*dimY)*k + (i2)*dimY + (j)] - Dx[(dimX*dimY)*k + (i)*dimY + (j)] + Dy[(dimX*dimY)*k + (i)*dimY + (j2)] - Dy[(dimX*dimY)*k + (i)*dimY + (j)] + Dz[(dimX*dimY)*k2 + (i)*dimY + (j)] - Dz[(dimX*dimY)*k + (i)*dimY + (j)];
+                b_val = -Bx[(dimX*dimY)*k + (i2)*dimY + (j)] + Bx[(dimX*dimY)*k + (i)*dimY + (j)] - By[(dimX*dimY)*k + (i)*dimY + (j2)] + By[(dimX*dimY)*k + (i)*dimY + (j)] - Bz[(dimX*dimY)*k2 + (i)*dimY + (j)] + Bz[(dimX*dimY)*k + (i)*dimY + (j)];
+                sum =  d_val + b_val;
+                sum += U[(dimX*dimY)*k + (i1)*dimY + (j)] + U[(dimX*dimY)*k + (i2)*dimY + (j)] + U[(dimX*dimY)*k + (i)*dimY + (j1)] + U[(dimX*dimY)*k + (i)*dimY + (j2)] + U[(dimX*dimY)*k1 + (i)*dimY + (j)] + U[(dimX*dimY)*k2 + (i)*dimY + (j)];
+                sum *= lambda;
+                sum += mu*A[(dimX*dimY)*k + (i)*dimY + (j)];
+                U[(dimX*dimY)*k + (i)*dimY + (j)] = normConst*sum;
+            }}}
+    return *U;
+}
+
+float updDxDyDz_shrinkAniso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda)
+{
+    int i,j,i1,j1,k,k1,index;
+    float val1, val11, val2, val22, val3, val33, denom_lam;
+    denom_lam = 1.0f/lambda;
+#pragma omp parallel for shared(U,denom_lam) private(index,i,j,i1,j1,k,k1,val1,val11,val2,val22,val3,val33)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            for(k=0; k<dimZ; k++) {
+                index = (dimX*dimY)*k + (i)*dimY + (j);
+                /* symmetric boundary conditions (Neuman) */
+                i1 = i+1; if (i1 == dimX) i1 = i-1;
+                j1 = j+1; if (j1 == dimY) j1 = j-1;
+                k1 = k+1; if (k1 == dimZ) k1 = k-1;
+                
+                val1 = (U[(dimX*dimY)*k + (i1)*dimY + (j)] - U[index]) + Bx[index];
+                val2 = (U[(dimX*dimY)*k + (i)*dimY + (j1)] - U[index]) + By[index];
+                val3 = (U[(dimX*dimY)*k1 + (i)*dimY + (j)] - U[index]) + Bz[index];
+                
+                val11 = fabs(val1) - denom_lam; if (val11 < 0) val11 = 0;
+                val22 = fabs(val2) - denom_lam; if (val22 < 0) val22 = 0;
+                val33 = fabs(val3) - denom_lam; if (val33 < 0) val33 = 0;
+                
+                if (val1 !=0) Dx[index] = (val1/fabs(val1))*val11; else Dx[index] = 0;
+                if (val2 !=0) Dy[index] = (val2/fabs(val2))*val22; else Dy[index] = 0;
+                if (val3 !=0) Dz[index] = (val3/fabs(val3))*val33; else Dz[index] = 0;
+                
+            }}}
+    return 1;
+}
+float updDxDyDz_shrinkIso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda)
+{
+    int i,j,i1,j1,k,k1,index;
+    float val1, val11, val2, val3, denom, denom_lam;
+    denom_lam = 1.0f/lambda;
+#pragma omp parallel for shared(U,denom_lam) private(index,denom,i,j,i1,j1,k,k1,val1,val11,val2,val3)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            for(k=0; k<dimZ; k++) {
+                index = (dimX*dimY)*k + (i)*dimY + (j);
+                /* symmetric boundary conditions (Neuman) */
+                i1 = i+1; if (i1 == dimX) i1 = i-1;
+                j1 = j+1; if (j1 == dimY) j1 = j-1;
+                k1 = k+1; if (k1 == dimZ) k1 = k-1;
+                
+                val1 = (U[(dimX*dimY)*k + (i1)*dimY + (j)] - U[index]) + Bx[index];
+                val2 = (U[(dimX*dimY)*k + (i)*dimY + (j1)] - U[index]) + By[index];
+                val3 = (U[(dimX*dimY)*k1 + (i)*dimY + (j)] - U[index]) + Bz[index];
+                
+                denom = sqrt(val1*val1 + val2*val2 + val3*val3);
+                
+                val11 = (denom - denom_lam); if (val11 < 0) val11 = 0.0f;
+                
+                if (denom != 0.0f) {
+                    Dx[index] = val11*(val1/denom);
+                    Dy[index] = val11*(val2/denom);
+                    Dz[index] = val11*(val3/denom);
+                }
+                else {
+                    Dx[index] = 0;
+                    Dy[index] = 0;
+                    Dz[index] = 0;
+                }               
+            }}}
+    return 1;
+}
+float updBxByBz3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ)
+{
+    int i,j,k,i1,j1,k1;
+#pragma omp parallel for shared(U) private(i,j,k,i1,j1,k1)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            for(k=0; k<dimZ; k++) {
+                /* symmetric boundary conditions (Neuman) */
+                i1 = i+1; if (i1 == dimX) i1 = i-1;
+                j1 = j+1; if (j1 == dimY) j1 = j-1;
+                k1 = k+1; if (k1 == dimZ) k1 = k-1;
+                
+                Bx[(dimX*dimY)*k + (i)*dimY + (j)] = Bx[(dimX*dimY)*k + (i)*dimY + (j)] + ((U[(dimX*dimY)*k + (i1)*dimY + (j)] - U[(dimX*dimY)*k + (i)*dimY + (j)]) - Dx[(dimX*dimY)*k + (i)*dimY + (j)]);
+                By[(dimX*dimY)*k + (i)*dimY + (j)] = By[(dimX*dimY)*k + (i)*dimY + (j)] + ((U[(dimX*dimY)*k + (i)*dimY + (j1)] - U[(dimX*dimY)*k + (i)*dimY + (j)]) - Dy[(dimX*dimY)*k + (i)*dimY + (j)]);
+                Bz[(dimX*dimY)*k + (i)*dimY + (j)] = Bz[(dimX*dimY)*k + (i)*dimY + (j)] + ((U[(dimX*dimY)*k1 + (i)*dimY + (j)] - U[(dimX*dimY)*k + (i)*dimY + (j)]) - Dz[(dimX*dimY)*k + (i)*dimY + (j)]);
+                
+            }}}
+    return 1;
+}
+/* General Functions */
+/*****************************************************************/
+/* Copy Image */
+float copyIm(float *A, float *B, int dimX, int dimY, int dimZ)
+{
+    int j;
+#pragma omp parallel for shared(A, B) private(j)
+    for(j=0; j<dimX*dimY*dimZ; j++)  B[j] = A[j];
+    return *B;
+}
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/SplitBregman_TV_core.h b/main_func/regularizers_CPU/SplitBregman_TV_core.h
new file mode 100644
index 0000000..ed9112c
--- /dev/null
+++ b/main_func/regularizers_CPU/SplitBregman_TV_core.h
@@ -0,0 +1,59 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+#include <matrix.h>
+#include <math.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <stdio.h>
+#include "omp.h"
+
+/* C-OMP implementation of Split Bregman - TV denoising-regularization model (2D/3D)
+*
+* Input Parameters:
+* 1. Noisy image/volume
+* 2. lambda - regularization parameter
+* 3. Number of iterations [OPTIONAL parameter]
+* 4. eplsilon - tolerance constant [OPTIONAL parameter]
+* 5. TV-type: 'iso' or 'l1' [OPTIONAL parameter]
+*
+* Output:
+* Filtered/regularized image
+*
+* Example:
+* figure;
+* Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+* u0 = Im + .05*randn(size(Im)); u0(u0 < 0) = 0;
+* u = SplitBregman_TV(single(u0), 10, 30, 1e-04);
+*
+* to compile with OMP support: mex SplitBregman_TV.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
+* References:
+* The Split Bregman Method for L1 Regularized Problems, by Tom Goldstein and Stanley Osher.
+* D. Kazantsev, 2016*
+*/
+
+float copyIm(float *A, float *B, int dimX, int dimY, int dimZ);
+float gauss_seidel2D(float *U, float *A, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda, float mu);
+float updDxDy_shrinkAniso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda);
+float updDxDy_shrinkIso2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY, float lambda);
+float updBxBy2D(float *U, float *Dx, float *Dy, float *Bx, float *By, int dimX, int dimY);
+
+float gauss_seidel3D(float *U, float *A, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda, float mu);
+float updDxDyDz_shrinkAniso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda);
+float updDxDyDz_shrinkIso3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ, float lambda);
+float updBxByBz3D(float *U, float *Dx, float *Dy, float *Dz, float *Bx, float *By, float *Bz, int dimX, int dimY, int dimZ);
diff --git a/main_func/regularizers_CPU/TGV_PD.c b/main_func/regularizers_CPU/TGV_PD.c
index 41f8615..8bce18a 100644
--- a/main_func/regularizers_CPU/TGV_PD.c
+++ b/main_func/regularizers_CPU/TGV_PD.c
@@ -1,10 +1,24 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include "TGV_PD_core.h"
 #include "mex.h"
-#include <matrix.h>
-#include <math.h>
-#include <stdlib.h>
-#include <memory.h>
-#include <stdio.h>
-#include "omp.h"
 
 /* C-OMP implementation of Primal-Dual denoising method for 
  * Total Generilized Variation (TGV)-L2 model (2D case only)
@@ -32,19 +46,6 @@
  * 28.11.16/Harwell
  */
  
-/* 2D functions */
-float DualP_2D(float *U, float *V1, float *V2, float *P1, float *P2, int dimX, int dimY, int dimZ, float sigma);
-float ProjP_2D(float *P1, float *P2, int dimX, int dimY, int dimZ, float alpha1);
-float DualQ_2D(float *V1, float *V2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float sigma);
-float ProjQ_2D(float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float alpha0);
-float DivProjP_2D(float *U, float *A, float *P1, float *P2, int dimX, int dimY, int dimZ, float lambda, float tau);
-float UpdV_2D(float *V1, float *V2, float *P1, float *P2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float tau);
-/*3D functions*/
-float DualP_3D(float *U, float *V1, float *V2, float *V3, float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ, float sigma);
-
-float newU(float *U, float *U_old, int dimX, int dimY, int dimZ);
-float copyIm(float *A, float *U, int dimX, int dimY, int dimZ);
-
 void mexFunction(
         int nlhs, mxArray *plhs[],
         int nrhs, const mxArray *prhs[])
@@ -183,171 +184,3 @@ void mexFunction(
 //         } /*end of iterations*/
 //     }
 }
-
-/*Calculating dual variable P (using forward differences)*/
-float DualP_2D(float *U, float *V1, float *V2, float *P1, float *P2, int dimX, int dimY, int dimZ, float sigma)
-{
-    int i,j;
-#pragma omp parallel for shared(U,V1,V2,P1,P2) private(i,j)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            if (i == dimX-1) P1[i*dimY + (j)] = P1[i*dimY + (j)] + sigma*((U[(i-1)*dimY + (j)] - U[i*dimY + (j)])  - V1[i*dimY + (j)]);
-            else P1[i*dimY + (j)] = P1[i*dimY + (j)] + sigma*((U[(i + 1)*dimY + (j)] - U[i*dimY + (j)])  - V1[i*dimY + (j)]);
-            if (j == dimY-1) P2[i*dimY + (j)] = P2[i*dimY + (j)] + sigma*((U[(i)*dimY + (j-1)] - U[i*dimY + (j)])  - V2[i*dimY + (j)]);
-            else  P2[i*dimY + (j)] = P2[i*dimY + (j)] + sigma*((U[(i)*dimY + (j+1)] - U[i*dimY + (j)])  - V2[i*dimY + (j)]);
-        }}
-    return 1;
-}
-/*Projection onto convex set for P*/
-float ProjP_2D(float *P1, float *P2, int dimX, int dimY, int dimZ, float alpha1)
-{
-    float grad_magn;
-    int i,j;
-#pragma omp parallel for shared(P1,P2) private(i,j,grad_magn)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            grad_magn = sqrt(pow(P1[i*dimY + (j)],2) + pow(P2[i*dimY + (j)],2));
-            grad_magn = grad_magn/alpha1;
-            if (grad_magn > 1.0) {
-                P1[i*dimY + (j)] = P1[i*dimY + (j)]/grad_magn;
-                P2[i*dimY + (j)] = P2[i*dimY + (j)]/grad_magn;
-            }
-        }}
-    return 1;
-}
-/*Calculating dual variable Q (using forward differences)*/
-float DualQ_2D(float *V1, float *V2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float sigma)
-{
-    int i,j;
-    float q1, q2, q11, q22;
-#pragma omp parallel for shared(Q1,Q2,Q3,V1,V2) private(i,j,q1,q2,q11,q22)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            if (i == dimX-1)
-            { q1 = (V1[(i-1)*dimY + (j)] - V1[i*dimY + (j)]);
-              q11 = (V2[(i-1)*dimY + (j)] - V2[i*dimY + (j)]);
-            }
-            else {
-                q1 = (V1[(i+1)*dimY + (j)] - V1[i*dimY + (j)]);
-                q11 = (V2[(i+1)*dimY + (j)] - V2[i*dimY + (j)]);
-            }
-            if (j == dimY-1) {
-                q2 = (V2[(i)*dimY + (j-1)] - V2[i*dimY + (j)]);
-                q22 = (V1[(i)*dimY + (j-1)] - V1[i*dimY + (j)]);
-            }
-            else {
-                q2 = (V2[(i)*dimY + (j+1)] - V2[i*dimY + (j)]);
-                q22 = (V1[(i)*dimY + (j+1)] - V1[i*dimY + (j)]);
-            }
-            Q1[i*dimY + (j)] = Q1[i*dimY + (j)] + sigma*(q1);
-            Q2[i*dimY + (j)] = Q2[i*dimY + (j)] + sigma*(q2);
-            Q3[i*dimY + (j)] = Q3[i*dimY + (j)]  + sigma*(0.5f*(q11 + q22));
-        }}
-    return 1;
-}
-
-float ProjQ_2D(float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float alpha0)
-{
-    float grad_magn;
-    int i,j;
-#pragma omp parallel for shared(Q1,Q2,Q3) private(i,j,grad_magn)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            grad_magn = sqrt(pow(Q1[i*dimY + (j)],2) + pow(Q2[i*dimY + (j)],2) + 2*pow(Q3[i*dimY + (j)],2));
-            grad_magn = grad_magn/alpha0;
-            if (grad_magn > 1.0) {
-                Q1[i*dimY + (j)] = Q1[i*dimY + (j)]/grad_magn;
-                Q2[i*dimY + (j)] = Q2[i*dimY + (j)]/grad_magn;
-                Q3[i*dimY + (j)] = Q3[i*dimY + (j)]/grad_magn;
-            }
-        }}
-    return 1;
-}
-/* Divergence and projection for P*/
-float DivProjP_2D(float *U, float *A, float *P1, float *P2, int dimX, int dimY, int dimZ, float lambda, float tau)
-{
-    int i,j;
-    float P_v1, P_v2, div;
-#pragma omp parallel for shared(U,A,P1,P2) private(i,j,P_v1,P_v2,div)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            if (i == 0) P_v1 = (P1[i*dimY + (j)]);
-            else P_v1 = (P1[i*dimY + (j)] - P1[(i-1)*dimY + (j)]);
-            if (j == 0) P_v2 = (P2[i*dimY + (j)]);
-            else  P_v2 = (P2[i*dimY + (j)] - P2[(i)*dimY + (j-1)]);
-            div = P_v1 + P_v2;
-            U[i*dimY + (j)] = (lambda*(U[i*dimY + (j)] + tau*div) + tau*A[i*dimY + (j)])/(lambda + tau);
-        }}
-    return *U;
-}
-/*get updated solution U*/
-float newU(float *U, float *U_old, int dimX, int dimY, int dimZ)
-{
-    int i;
-#pragma omp parallel for shared(U,U_old) private(i)
-    for(i=0; i<dimX*dimY*dimZ; i++) U[i] = 2*U[i] - U_old[i];
-    return *U;
-}
-
-/*get update for V*/
-float UpdV_2D(float *V1, float *V2, float *P1, float *P2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float tau)
-{
-    int i,j;
-    float q1, q11, q2, q22, div1, div2;
-#pragma omp parallel for shared(V1,V2,P1,P2,Q1,Q2,Q3) private(i,j, q1, q11, q2, q22, div1, div2)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            /* symmetric boundary conditions (Neuman) */
-            if (i == 0) {
-                q1 = (Q1[i*dimY + (j)]);
-                q11 = (Q3[i*dimY + (j)]);
-            }
-            else {
-                q1 = (Q1[i*dimY + (j)] - Q1[(i-1)*dimY + (j)]);
-                q11 = (Q3[i*dimY + (j)] - Q3[(i-1)*dimY + (j)]);
-            }
-            if (j == 0) {
-                q2 = (Q2[i*dimY + (j)]);
-                q22 = (Q3[i*dimY + (j)]);
-            }
-            else  {
-                q2 = (Q2[i*dimY + (j)] - Q2[(i)*dimY + (j-1)]);
-                q22 = (Q3[i*dimY + (j)] - Q3[(i)*dimY + (j-1)]);
-            }
-            div1 = q1 + q22;
-            div2 = q2 + q11;
-            V1[i*dimY + (j)] = V1[i*dimY + (j)] + tau*(P1[i*dimY + (j)] + div1);
-            V2[i*dimY + (j)] = V2[i*dimY + (j)] + tau*(P2[i*dimY + (j)] + div2);
-        }}
-    return 1;
-}
-/* Copy Image */
-float copyIm(float *A, float *U, int dimX, int dimY, int dimZ)
-{
-    int j;
-#pragma omp parallel for shared(A, U) private(j)
-    for(j=0; j<dimX*dimY*dimZ; j++)  U[j] = A[j];
-    return *U;
-}
-/*********************3D *********************/
-
-/*Calculating dual variable P (using forward differences)*/
-float DualP_3D(float *U, float *V1, float *V2, float *V3, float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ, float sigma)
-{
-    int i,j,k;
-#pragma omp parallel for shared(U,V1,V2,V3,P1,P2,P3) private(i,j,k)
-    for(i=0; i<dimX; i++) {
-        for(j=0; j<dimY; j++) {
-            for(k=0; k<dimZ; k++) {
-                /* symmetric boundary conditions (Neuman) */
-                if (i == dimX-1) P1[dimX*dimY*k + i*dimY + (j)] = P1[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i-1)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V1[dimX*dimY*k + i*dimY + (j)]);
-                else P1[dimX*dimY*k + i*dimY + (j)] = P1[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i + 1)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V1[dimX*dimY*k + i*dimY + (j)]);
-                if (j == dimY-1) P2[dimX*dimY*k + i*dimY + (j)] = P2[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i)*dimY + (j-1)] - U[dimX*dimY*k + i*dimY + (j)])  - V2[dimX*dimY*k + i*dimY + (j)]);
-                else  P2[dimX*dimY*k + i*dimY + (j)] = P2[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i)*dimY + (j+1)] - U[dimX*dimY*k + i*dimY + (j)])  - V2[dimX*dimY*k + i*dimY + (j)]);
-                if (k == dimZ-1) P3[dimX*dimY*k + i*dimY + (j)] = P3[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*(k-1) + (i)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V3[dimX*dimY*k + i*dimY + (j)]);
-                else  P3[dimX*dimY*k + i*dimY + (j)] = P3[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*(k+1) + (i)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V3[dimX*dimY*k + i*dimY + (j)]);
-            }}}
-    return 1;
-}
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/TGV_PD_core.c b/main_func/regularizers_CPU/TGV_PD_core.c
new file mode 100644
index 0000000..4c0427c
--- /dev/null
+++ b/main_func/regularizers_CPU/TGV_PD_core.c
@@ -0,0 +1,217 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include "TGV_PD_core.h"
+
+/* C-OMP implementation of Primal-Dual denoising method for 
+ * Total Generilized Variation (TGV)-L2 model (2D case only)
+ *
+ * Input Parameters:
+ * 1. Noisy image/volume (2D)
+ * 2. lambda - regularization parameter
+ * 3. parameter to control first-order term (alpha1)
+ * 4. parameter to control the second-order term (alpha0)
+ * 5. Number of CP iterations
+ *
+ * Output:
+ * Filtered/regularized image 
+ *
+ * Example:
+ * figure;
+ * Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+ * u0 = Im + .03*randn(size(Im)); % adding noise
+ * tic; u = PrimalDual_TGV(single(u0), 0.02, 1.3, 1, 550); toc;
+ *
+ * to compile with OMP support: mex TGV_PD.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
+ * References:
+ * K. Bredies "Total Generalized Variation"
+ *
+ * 28.11.16/Harwell
+ */
+ 
+
+
+
+/*Calculating dual variable P (using forward differences)*/
+float DualP_2D(float *U, float *V1, float *V2, float *P1, float *P2, int dimX, int dimY, int dimZ, float sigma)
+{
+    int i,j;
+#pragma omp parallel for shared(U,V1,V2,P1,P2) private(i,j)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            if (i == dimX-1) P1[i*dimY + (j)] = P1[i*dimY + (j)] + sigma*((U[(i-1)*dimY + (j)] - U[i*dimY + (j)])  - V1[i*dimY + (j)]);
+            else P1[i*dimY + (j)] = P1[i*dimY + (j)] + sigma*((U[(i + 1)*dimY + (j)] - U[i*dimY + (j)])  - V1[i*dimY + (j)]);
+            if (j == dimY-1) P2[i*dimY + (j)] = P2[i*dimY + (j)] + sigma*((U[(i)*dimY + (j-1)] - U[i*dimY + (j)])  - V2[i*dimY + (j)]);
+            else  P2[i*dimY + (j)] = P2[i*dimY + (j)] + sigma*((U[(i)*dimY + (j+1)] - U[i*dimY + (j)])  - V2[i*dimY + (j)]);
+        }}
+    return 1;
+}
+/*Projection onto convex set for P*/
+float ProjP_2D(float *P1, float *P2, int dimX, int dimY, int dimZ, float alpha1)
+{
+    float grad_magn;
+    int i,j;
+#pragma omp parallel for shared(P1,P2) private(i,j,grad_magn)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            grad_magn = sqrt(pow(P1[i*dimY + (j)],2) + pow(P2[i*dimY + (j)],2));
+            grad_magn = grad_magn/alpha1;
+            if (grad_magn > 1.0) {
+                P1[i*dimY + (j)] = P1[i*dimY + (j)]/grad_magn;
+                P2[i*dimY + (j)] = P2[i*dimY + (j)]/grad_magn;
+            }
+        }}
+    return 1;
+}
+/*Calculating dual variable Q (using forward differences)*/
+float DualQ_2D(float *V1, float *V2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float sigma)
+{
+    int i,j;
+    float q1, q2, q11, q22;
+#pragma omp parallel for shared(Q1,Q2,Q3,V1,V2) private(i,j,q1,q2,q11,q22)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            if (i == dimX-1)
+            { q1 = (V1[(i-1)*dimY + (j)] - V1[i*dimY + (j)]);
+              q11 = (V2[(i-1)*dimY + (j)] - V2[i*dimY + (j)]);
+            }
+            else {
+                q1 = (V1[(i+1)*dimY + (j)] - V1[i*dimY + (j)]);
+                q11 = (V2[(i+1)*dimY + (j)] - V2[i*dimY + (j)]);
+            }
+            if (j == dimY-1) {
+                q2 = (V2[(i)*dimY + (j-1)] - V2[i*dimY + (j)]);
+                q22 = (V1[(i)*dimY + (j-1)] - V1[i*dimY + (j)]);
+            }
+            else {
+                q2 = (V2[(i)*dimY + (j+1)] - V2[i*dimY + (j)]);
+                q22 = (V1[(i)*dimY + (j+1)] - V1[i*dimY + (j)]);
+            }
+            Q1[i*dimY + (j)] = Q1[i*dimY + (j)] + sigma*(q1);
+            Q2[i*dimY + (j)] = Q2[i*dimY + (j)] + sigma*(q2);
+            Q3[i*dimY + (j)] = Q3[i*dimY + (j)]  + sigma*(0.5f*(q11 + q22));
+        }}
+    return 1;
+}
+
+float ProjQ_2D(float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float alpha0)
+{
+    float grad_magn;
+    int i,j;
+#pragma omp parallel for shared(Q1,Q2,Q3) private(i,j,grad_magn)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            grad_magn = sqrt(pow(Q1[i*dimY + (j)],2) + pow(Q2[i*dimY + (j)],2) + 2*pow(Q3[i*dimY + (j)],2));
+            grad_magn = grad_magn/alpha0;
+            if (grad_magn > 1.0) {
+                Q1[i*dimY + (j)] = Q1[i*dimY + (j)]/grad_magn;
+                Q2[i*dimY + (j)] = Q2[i*dimY + (j)]/grad_magn;
+                Q3[i*dimY + (j)] = Q3[i*dimY + (j)]/grad_magn;
+            }
+        }}
+    return 1;
+}
+/* Divergence and projection for P*/
+float DivProjP_2D(float *U, float *A, float *P1, float *P2, int dimX, int dimY, int dimZ, float lambda, float tau)
+{
+    int i,j;
+    float P_v1, P_v2, div;
+#pragma omp parallel for shared(U,A,P1,P2) private(i,j,P_v1,P_v2,div)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            if (i == 0) P_v1 = (P1[i*dimY + (j)]);
+            else P_v1 = (P1[i*dimY + (j)] - P1[(i-1)*dimY + (j)]);
+            if (j == 0) P_v2 = (P2[i*dimY + (j)]);
+            else  P_v2 = (P2[i*dimY + (j)] - P2[(i)*dimY + (j-1)]);
+            div = P_v1 + P_v2;
+            U[i*dimY + (j)] = (lambda*(U[i*dimY + (j)] + tau*div) + tau*A[i*dimY + (j)])/(lambda + tau);
+        }}
+    return *U;
+}
+/*get updated solution U*/
+float newU(float *U, float *U_old, int dimX, int dimY, int dimZ)
+{
+    int i;
+#pragma omp parallel for shared(U,U_old) private(i)
+    for(i=0; i<dimX*dimY*dimZ; i++) U[i] = 2*U[i] - U_old[i];
+    return *U;
+}
+
+/*get update for V*/
+float UpdV_2D(float *V1, float *V2, float *P1, float *P2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float tau)
+{
+    int i,j;
+    float q1, q11, q2, q22, div1, div2;
+#pragma omp parallel for shared(V1,V2,P1,P2,Q1,Q2,Q3) private(i,j, q1, q11, q2, q22, div1, div2)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            /* symmetric boundary conditions (Neuman) */
+            if (i == 0) {
+                q1 = (Q1[i*dimY + (j)]);
+                q11 = (Q3[i*dimY + (j)]);
+            }
+            else {
+                q1 = (Q1[i*dimY + (j)] - Q1[(i-1)*dimY + (j)]);
+                q11 = (Q3[i*dimY + (j)] - Q3[(i-1)*dimY + (j)]);
+            }
+            if (j == 0) {
+                q2 = (Q2[i*dimY + (j)]);
+                q22 = (Q3[i*dimY + (j)]);
+            }
+            else  {
+                q2 = (Q2[i*dimY + (j)] - Q2[(i)*dimY + (j-1)]);
+                q22 = (Q3[i*dimY + (j)] - Q3[(i)*dimY + (j-1)]);
+            }
+            div1 = q1 + q22;
+            div2 = q2 + q11;
+            V1[i*dimY + (j)] = V1[i*dimY + (j)] + tau*(P1[i*dimY + (j)] + div1);
+            V2[i*dimY + (j)] = V2[i*dimY + (j)] + tau*(P2[i*dimY + (j)] + div2);
+        }}
+    return 1;
+}
+/* Copy Image */
+float copyIm(float *A, float *U, int dimX, int dimY, int dimZ)
+{
+    int j;
+#pragma omp parallel for shared(A, U) private(j)
+    for(j=0; j<dimX*dimY*dimZ; j++)  U[j] = A[j];
+    return *U;
+}
+/*********************3D *********************/
+
+/*Calculating dual variable P (using forward differences)*/
+float DualP_3D(float *U, float *V1, float *V2, float *V3, float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ, float sigma)
+{
+    int i,j,k;
+#pragma omp parallel for shared(U,V1,V2,V3,P1,P2,P3) private(i,j,k)
+    for(i=0; i<dimX; i++) {
+        for(j=0; j<dimY; j++) {
+            for(k=0; k<dimZ; k++) {
+                /* symmetric boundary conditions (Neuman) */
+                if (i == dimX-1) P1[dimX*dimY*k + i*dimY + (j)] = P1[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i-1)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V1[dimX*dimY*k + i*dimY + (j)]);
+                else P1[dimX*dimY*k + i*dimY + (j)] = P1[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i + 1)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V1[dimX*dimY*k + i*dimY + (j)]);
+                if (j == dimY-1) P2[dimX*dimY*k + i*dimY + (j)] = P2[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i)*dimY + (j-1)] - U[dimX*dimY*k + i*dimY + (j)])  - V2[dimX*dimY*k + i*dimY + (j)]);
+                else  P2[dimX*dimY*k + i*dimY + (j)] = P2[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*k + (i)*dimY + (j+1)] - U[dimX*dimY*k + i*dimY + (j)])  - V2[dimX*dimY*k + i*dimY + (j)]);
+                if (k == dimZ-1) P3[dimX*dimY*k + i*dimY + (j)] = P3[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*(k-1) + (i)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V3[dimX*dimY*k + i*dimY + (j)]);
+                else  P3[dimX*dimY*k + i*dimY + (j)] = P3[dimX*dimY*k + i*dimY + (j)] + sigma*((U[dimX*dimY*(k+1) + (i)*dimY + (j)] - U[dimX*dimY*k + i*dimY + (j)])  - V3[dimX*dimY*k + i*dimY + (j)]);
+            }}}
+    return 1;
+}
\ No newline at end of file
diff --git a/main_func/regularizers_CPU/TGV_PD_core.h b/main_func/regularizers_CPU/TGV_PD_core.h
new file mode 100644
index 0000000..e25ae68
--- /dev/null
+++ b/main_func/regularizers_CPU/TGV_PD_core.h
@@ -0,0 +1,64 @@
+/*
+This work is part of the Core Imaging Library developed by
+Visual Analytics and Imaging System Group of the Science Technology
+Facilities Council, STFC
+
+Copyright 2017 Daniil Kazanteev
+Copyright 2017 Srikanth Nagella, Edoardo Pasca
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+#include <matrix.h>
+#include <math.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <stdio.h>
+#include "omp.h"
+
+/* C-OMP implementation of Primal-Dual denoising method for
+* Total Generilized Variation (TGV)-L2 model (2D case only)
+*
+* Input Parameters:
+* 1. Noisy image/volume (2D)
+* 2. lambda - regularization parameter
+* 3. parameter to control first-order term (alpha1)
+* 4. parameter to control the second-order term (alpha0)
+* 5. Number of CP iterations
+*
+* Output:
+* Filtered/regularized image
+*
+* Example:
+* figure;
+* Im = double(imread('lena_gray_256.tif'))/255;  % loading image
+* u0 = Im + .03*randn(size(Im)); % adding noise
+* tic; u = PrimalDual_TGV(single(u0), 0.02, 1.3, 1, 550); toc;
+*
+* to compile with OMP support: mex TGV_PD.c CFLAGS="\$CFLAGS -fopenmp -Wall -std=c99" LDFLAGS="\$LDFLAGS -fopenmp"
+* References:
+* K. Bredies "Total Generalized Variation"
+*
+* 28.11.16/Harwell
+*/
+
+/* 2D functions */
+float DualP_2D(float *U, float *V1, float *V2, float *P1, float *P2, int dimX, int dimY, int dimZ, float sigma);
+float ProjP_2D(float *P1, float *P2, int dimX, int dimY, int dimZ, float alpha1);
+float DualQ_2D(float *V1, float *V2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float sigma);
+float ProjQ_2D(float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float alpha0);
+float DivProjP_2D(float *U, float *A, float *P1, float *P2, int dimX, int dimY, int dimZ, float lambda, float tau);
+float UpdV_2D(float *V1, float *V2, float *P1, float *P2, float *Q1, float *Q2, float *Q3, int dimX, int dimY, int dimZ, float tau);
+/*3D functions*/
+float DualP_3D(float *U, float *V1, float *V2, float *V3, float *P1, float *P2, float *P3, int dimX, int dimY, int dimZ, float sigma);
+
+float newU(float *U, float *U_old, int dimX, int dimY, int dimZ);
+float copyIm(float *A, float *U, int dimX, int dimY, int dimZ);