diff --git a/modules/core/include/opencv2/core/ocl.hpp b/modules/core/include/opencv2/core/ocl.hpp
index f50ed378f9..f2535940f0 100644
--- a/modules/core/include/opencv2/core/ocl.hpp
+++ b/modules/core/include/opencv2/core/ocl.hpp
@@ -286,7 +286,7 @@ public:
     Kernel();
     Kernel(const char* kname, const Program& prog);
     Kernel(const char* kname, const ProgramSource2& prog,
-           const String& buildopts, String* errmsg=0);
+           const String& buildopts = String(), String* errmsg=0);
     ~Kernel();
     Kernel(const Kernel& k);
     Kernel& operator = (const Kernel& k);
diff --git a/modules/core/src/ocl.cpp b/modules/core/src/ocl.cpp
index 64460efb0d..8a0cc279a6 100644
--- a/modules/core/src/ocl.cpp
+++ b/modules/core/src/ocl.cpp
@@ -1893,7 +1893,7 @@ Context2& Context2::getDefault()
         // First, try to retrieve existing context of the same type.
         // In its turn, Platform::getContext() may call Context2::create()
         // if there is no such context.
-        ctx.create(Device::TYPE_ACCELERATOR);
+        ctx.create(Device::TYPE_CPU);
         if(!ctx.p)
             ctx.create(Device::TYPE_DGPU);
         if(!ctx.p)
@@ -2041,6 +2041,7 @@ struct Kernel::Impl
         cl_int retval = 0;
         handle = ph != 0 ?
             clCreateKernel(ph, kname, &retval) : 0;
+        printf("kernel creation error code: %d\n", retval);
         for( int i = 0; i < MAX_ARRS; i++ )
             u[i] = 0;
         haveTempDstUMats = false;
@@ -2218,7 +2219,7 @@ int Kernel::set(int i, const KernelArg& arg)
         else if( arg.m->dims <= 2 )
         {
             UMat2D u2d(*arg.m);
-            clSetKernelArg(p->handle, (cl_uint)i, sizeof(h), &h);
+            clSetKernelArg(p->handle, (cl_uint)i, sizeof(h), &h));
             clSetKernelArg(p->handle, (cl_uint)(i+1), sizeof(u2d.step), &u2d.step);
             clSetKernelArg(p->handle, (cl_uint)(i+2), sizeof(u2d.offset), &u2d.offset);
             i += 3;
diff --git a/modules/imgproc/src/imgwarp.cpp b/modules/imgproc/src/imgwarp.cpp
index 15d7c6a875..e5383be04b 100644
--- a/modules/imgproc/src/imgwarp.cpp
+++ b/modules/imgproc/src/imgwarp.cpp
@@ -4030,16 +4030,76 @@ private:
 };
 #endif
 
+static bool ocl_warpPerspective(InputArray _src, OutputArray _dst, InputArray _M0,
+                                Size dsize, int flags, int borderType, const Scalar& borderValue)
+{
+    int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type), wdepth = depth;
+    double doubleSupport = ocl::Device::getDefault().doubleFPConfig() > 0;
+
+    int interpolation = flags & INTER_MAX;
+    if( interpolation == INTER_AREA )
+        interpolation = INTER_LINEAR;
+
+    if ( !(borderType == cv::BORDER_CONSTANT &&
+           (interpolation == cv::INTER_NEAREST || interpolation == cv::INTER_LINEAR || interpolation == cv::INTER_CUBIC)) ||
+         (!doubleSupport && depth == CV_64F) || cn > 4 || cn == 3)
+        return false;
+
+    UMat src = _src.getUMat(), M0;
+    _dst.create( dsize.area() == 0 ? src.size() : dsize, src.type() );
+    UMat dst = _dst.getUMat();
+
+    double M[9];
+    Mat matM(3, 3, doubleSupport ? CV_64F : CV_32F, M), M1 = _M0.getMat();
+    CV_Assert( (M1.type() == CV_32F || M1.type() == CV_64F) && M1.rows == 3 && M1.cols == 3 );
+    M1.convertTo(matM, matM.type());
+    if( !(flags & WARP_INVERSE_MAP) )
+         invert(matM, matM);
+    matM.copyTo(M0);
+
+    const char * const interpolationMap[3] = { "NEAREST", "LINEAR", "CUBIC" };
+    ocl::Kernel k;
+
+    if (interpolation == INTER_NEAREST)
+    {
+        k.create("warpPerspective", ocl::imgproc::warp_perspective_oclsrc,
+                 format("-D INTER_NEAREST -D T=%s%s", ocl::typeToStr(type),
+                        doubleSupport ? " -D DOUBLE_SUPPORT" : ""));
+    }
+    else
+    {
+        char cvt[2][50];
+        wdepth = std::max(CV_32S, depth);
+        k.create("warpPerspective", ocl::imgproc::warp_perspective_oclsrc,
+                  format("-D INTER_%s -D T=%s -D WT=%s -D depth=%d -D convertToWT=%s -D convertToT=%s%s",
+                         interpolationMap[interpolation], ocl::typeToStr(type),
+                         ocl::typeToStr(CV_MAKE_TYPE(wdepth, cn)), depth,
+                         ocl::convertTypeStr(depth, wdepth, cn, cvt[0]),
+                         ocl::convertTypeStr(wdepth, depth, cn, cvt[1]),
+                         doubleSupport ? " -D DOUBLE_SUPPORT" : ""));
+    }
+
+    k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst),
+           ocl::KernelArg::PtrOnly(M0), ocl::KernelArg::Constant(Mat(1, 1, CV_MAKE_TYPE(wdepth, cn), borderValue)));
+
+    size_t globalThreads[2] = { dst.cols, dst.rows };
+    return k.run(2, globalThreads, NULL, false);
+}
+
 }
 
 void cv::warpPerspective( InputArray _src, OutputArray _dst, InputArray _M0,
                           Size dsize, int flags, int borderType, const Scalar& borderValue )
 {
+    CV_Assert( _src.total() > 0 );
+
+    if (ocl::useOpenCL() && _dst.isUMat() && ocl_warpPerspective(_src, _dst, _M0, dsize, flags, borderType, borderValue))
+        return;
+
     Mat src = _src.getMat(), M0 = _M0.getMat();
     _dst.create( dsize.area() == 0 ? src.size() : dsize, src.type() );
     Mat dst = _dst.getMat();
 
-    CV_Assert( src.cols > 0 && src.rows > 0 );
     if( dst.data == src.data )
         src = src.clone();
 
diff --git a/modules/imgproc/src/opencl/warp_affine.cl b/modules/imgproc/src/opencl/warp_affine.cl
new file mode 100644
index 0000000000..27f99e005e
--- /dev/null
+++ b/modules/imgproc/src/opencl/warp_affine.cl
@@ -0,0 +1,761 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                           License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//    Zhang Ying, zhangying913@gmail.com
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors as is and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+
+//warpAffine kernel
+//support data types: CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4, and three interpolation methods: NN, Linear, Cubic.
+
+#ifdef DOUBLE_SUPPORT
+#ifdef cl_amd_fp64
+#pragma OPENCL EXTENSION cl_amd_fp64:enable
+#elif defined (cl_khr_fp64)
+#pragma OPENCL EXTENSION cl_khr_fp64:enable
+#endif
+typedef double F;
+typedef double4 F4;
+#define convert_F4 convert_double4
+#else
+typedef float F;
+typedef float4 F4;
+#define convert_F4 convert_float4
+#endif
+
+#define INTER_BITS 5
+#define INTER_TAB_SIZE (1 << INTER_BITS)
+#define INTER_SCALE 1.f/INTER_TAB_SIZE
+#define AB_BITS max(10, (int)INTER_BITS)
+#define AB_SCALE (1 << AB_BITS)
+#define INTER_REMAP_COEF_BITS 15
+#define INTER_REMAP_COEF_SCALE (1 << INTER_REMAP_COEF_BITS)
+
+inline void interpolateCubic( float x, float* coeffs )
+{
+    const float A = -0.75f;
+
+    coeffs[0] = ((A*(x + 1.f) - 5.0f*A)*(x + 1.f) + 8.0f*A)*(x + 1.f) - 4.0f*A;
+    coeffs[1] = ((A + 2.f)*x - (A + 3.f))*x*x + 1.f;
+    coeffs[2] = ((A + 2.f)*(1.f - x) - (A + 3.f))*(1.f - x)*(1.f - x) + 1.f;
+    coeffs[3] = 1.f - coeffs[0] - coeffs[1] - coeffs[2];
+}
+
+
+/**********************************************8UC1*********************************************
+***********************************************************************************************/
+__kernel void warpAffineNN_C1_D0(__global uchar const * restrict src, __global uchar * dst, int src_cols, int src_rows,
+                                 int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                 int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        dx = (dx<<2) - (dst_offset&3);
+
+        int round_delta = (AB_SCALE>>1);
+
+        int4 X, Y;
+        int4 sx, sy;
+        int4 DX = (int4)(dx, dx+1, dx+2, dx+3);
+        DX = (DX << AB_BITS);
+        F4 M0DX, M3DX;
+        M0DX = M[0] * convert_F4(DX);
+        M3DX = M[3] * convert_F4(DX);
+        X = convert_int4(rint(M0DX));
+        Y = convert_int4(rint(M3DX));
+        int tmp1, tmp2;
+        tmp1 = rint((M[1]*dy + M[2]) * AB_SCALE);
+        tmp2 = rint((M[4]*dy + M[5]) * AB_SCALE);
+
+        X += tmp1 + round_delta;
+        Y += tmp2 + round_delta;
+
+        sx = convert_int4(convert_short4(X >> AB_BITS));
+        sy = convert_int4(convert_short4(Y >> AB_BITS));
+
+        __global uchar4 * d = (__global uchar4 *)(dst+dst_offset+dy*dstStep+dx);
+        uchar4 dval = *d;
+        DX = (int4)(dx, dx+1, dx+2, dx+3);
+        int4 dcon = DX >= 0 && DX < dst_cols && dy >= 0 && dy < dst_rows;
+        int4 scon = sx >= 0 && sx < src_cols && sy >= 0 && sy < src_rows;
+        int4 spos = src_offset + sy * srcStep + sx;
+        uchar4 sval;
+        sval.s0 = scon.s0 ? src[spos.s0] : 0;
+        sval.s1 = scon.s1 ? src[spos.s1] : 0;
+        sval.s2 = scon.s2 ? src[spos.s2] : 0;
+        sval.s3 = scon.s3 ? src[spos.s3] : 0;
+        dval = convert_uchar4(dcon) != (uchar4)(0,0,0,0) ? sval : dval;
+        *d = dval;
+    }
+}
+
+__kernel void warpAffineLinear_C1_D0(__global const uchar * restrict src, __global uchar * dst, int src_cols, int src_rows,
+                                     int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                     int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        dx = (dx<<2) - (dst_offset&3);
+
+        int round_delta = ((AB_SCALE >> INTER_BITS) >> 1);
+
+        int4 X, Y;
+        short4  ax, ay;
+        int4 sx, sy;
+        int4 DX = (int4)(dx, dx+1, dx+2, dx+3);
+        DX = (DX << AB_BITS);
+        F4 M0DX, M3DX;
+        M0DX = M[0] * convert_F4(DX);
+        M3DX = M[3] * convert_F4(DX);
+        X = convert_int4(rint(M0DX));
+        Y = convert_int4(rint(M3DX));
+
+        int tmp1, tmp2;
+        tmp1 = rint((M[1]*dy + M[2]) * AB_SCALE);
+        tmp2 = rint((M[4]*dy + M[5]) * AB_SCALE);
+
+        X += tmp1 + round_delta;
+        Y += tmp2 + round_delta;
+
+        X = X >> (AB_BITS - INTER_BITS);
+        Y = Y >> (AB_BITS - INTER_BITS);
+
+        sx = convert_int4(convert_short4(X >> INTER_BITS));
+        sy = convert_int4(convert_short4(Y >> INTER_BITS));
+        ax = convert_short4(X & (INTER_TAB_SIZE-1));
+        ay = convert_short4(Y & (INTER_TAB_SIZE-1));
+
+        uchar4 v0, v1, v2,v3;
+        int4 scon0, scon1, scon2, scon3;
+        int4 spos0, spos1, spos2, spos3;
+
+        scon0 = (sx >= 0 && sx < src_cols && sy >= 0 && sy < src_rows);
+        scon1 = (sx+1 >= 0 && sx+1 < src_cols && sy >= 0 && sy < src_rows);
+        scon2 = (sx >= 0 && sx < src_cols && sy+1 >= 0 && sy+1 < src_rows);
+        scon3 = (sx+1 >= 0 && sx+1 < src_cols && sy+1 >= 0 && sy+1 < src_rows);
+        spos0 = src_offset + sy * srcStep + sx;
+        spos1 = src_offset + sy * srcStep + sx + 1;
+        spos2 = src_offset + (sy+1) * srcStep + sx;
+        spos3 = src_offset + (sy+1) * srcStep + sx + 1;
+
+        v0.s0 = scon0.s0 ? src[spos0.s0] : 0;
+        v1.s0 = scon1.s0 ? src[spos1.s0] : 0;
+        v2.s0 = scon2.s0 ? src[spos2.s0] : 0;
+        v3.s0 = scon3.s0 ? src[spos3.s0] : 0;
+
+        v0.s1 = scon0.s1 ? src[spos0.s1] : 0;
+        v1.s1 = scon1.s1 ? src[spos1.s1] : 0;
+        v2.s1 = scon2.s1 ? src[spos2.s1] : 0;
+        v3.s1 = scon3.s1 ? src[spos3.s1] : 0;
+
+        v0.s2 = scon0.s2 ? src[spos0.s2] : 0;
+        v1.s2 = scon1.s2 ? src[spos1.s2] : 0;
+        v2.s2 = scon2.s2 ? src[spos2.s2] : 0;
+        v3.s2 = scon3.s2 ? src[spos3.s2] : 0;
+
+        v0.s3 = scon0.s3 ? src[spos0.s3] : 0;
+        v1.s3 = scon1.s3 ? src[spos1.s3] : 0;
+        v2.s3 = scon2.s3 ? src[spos2.s3] : 0;
+        v3.s3 = scon3.s3 ? src[spos3.s3] : 0;
+
+        short4 itab0, itab1, itab2, itab3;
+        float4 taby, tabx;
+        taby = INTER_SCALE * convert_float4(ay);
+        tabx = INTER_SCALE * convert_float4(ax);
+
+        itab0 = convert_short4_sat(( (1.0f-taby)*(1.0f-tabx) * (float4)INTER_REMAP_COEF_SCALE ));
+        itab1 = convert_short4_sat(( (1.0f-taby)*tabx * (float4)INTER_REMAP_COEF_SCALE ));
+        itab2 = convert_short4_sat(( taby*(1.0f-tabx) * (float4)INTER_REMAP_COEF_SCALE ));
+        itab3 = convert_short4_sat(( taby*tabx * (float4)INTER_REMAP_COEF_SCALE ));
+
+
+        int4 val;
+        uchar4 tval;
+        val = convert_int4(v0) * convert_int4(itab0) + convert_int4(v1) * convert_int4(itab1)
+              + convert_int4(v2) * convert_int4(itab2) + convert_int4(v3) * convert_int4(itab3);
+        tval = convert_uchar4_sat ( (val + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS ) ;
+
+        __global uchar4 * d =(__global uchar4 *)(dst+dst_offset+dy*dstStep+dx);
+        uchar4 dval = *d;
+        DX = (int4)(dx, dx+1, dx+2, dx+3);
+        int4 dcon = DX >= 0 && DX < dst_cols && dy >= 0 && dy < dst_rows;
+        dval = convert_uchar4(dcon != 0) ? tval : dval;
+        *d = dval;
+    }
+}
+
+__kernel void warpAffineCubic_C1_D0(__global uchar * src, __global uchar * dst, int src_cols, int src_rows,
+                                    int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                    int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = ((AB_SCALE>>INTER_BITS)>>1);
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        int X = X0 >> (AB_BITS - INTER_BITS);
+        int Y = Y0 >> (AB_BITS - INTER_BITS);
+
+        short sx = (short)(X >> INTER_BITS) - 1;
+        short sy = (short)(Y >> INTER_BITS) - 1;
+        short ay = (short)(Y & (INTER_TAB_SIZE-1));
+        short ax = (short)(X & (INTER_TAB_SIZE-1));
+
+        uchar v[16];
+        int i, j;
+
+#pragma unroll 4
+        for(i=0; i<4;  i++)
+            for(j=0; j<4;  j++)
+            {
+                v[i*4+j] = (sx+j >= 0 && sx+j < src_cols && sy+i >= 0 && sy+i < src_rows) ? src[src_offset+(sy+i) * srcStep + (sx+j)] : 0;
+            }
+
+        short itab[16];
+        float tab1y[4], tab1x[4];
+        float axx, ayy;
+
+        ayy = 1.f/INTER_TAB_SIZE * ay;
+        axx = 1.f/INTER_TAB_SIZE * ax;
+        interpolateCubic(ayy, tab1y);
+        interpolateCubic(axx, tab1x);
+        int isum = 0;
+
+#pragma unroll 16
+        for( i=0; i<16; i++ )
+        {
+            F v = tab1y[(i>>2)] * tab1x[(i&3)];
+            isum += itab[i] = convert_short_sat( rint( v * INTER_REMAP_COEF_SCALE ) );
+        }
+
+        if( isum != INTER_REMAP_COEF_SCALE )
+        {
+            int k1, k2;
+            int diff = isum - INTER_REMAP_COEF_SCALE;
+            int Mk1=2, Mk2=2, mk1=2, mk2=2;
+            for( k1 = 2; k1 < 4; k1++ )
+                for( k2 = 2; k2 < 4; k2++ )
+                {
+                    if( itab[(k1<<2)+k2] < itab[(mk1<<2)+mk2] )
+                        mk1 = k1, mk2 = k2;
+                    else if( itab[(k1<<2)+k2] > itab[(Mk1<<2)+Mk2] )
+                        Mk1 = k1, Mk2 = k2;
+                }
+            diff<0 ? (itab[(Mk1<<2)+Mk2]=(short)(itab[(Mk1<<2)+Mk2]-diff)) : (itab[(mk1<<2)+mk2]=(short)(itab[(mk1<<2)+mk2]-diff));
+        }
+
+        if( dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+        {
+            int sum=0;
+            for ( i =0; i<16; i++ )
+            {
+                sum += v[i] * itab[i] ;
+            }
+            dst[dst_offset+dy*dstStep+dx] = convert_uchar_sat( (sum + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS ) ;
+        }
+    }
+}
+
+/**********************************************8UC4*********************************************
+***********************************************************************************************/
+
+__kernel void warpAffineNN_C4_D0(__global uchar4 const * restrict src, __global uchar4 * dst, int src_cols, int src_rows,
+                                 int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                 int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = (AB_SCALE >> 1);
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+
+        int sx0 = (short)(X0 >> AB_BITS);
+        int sy0 = (short)(Y0 >> AB_BITS);
+
+        if(dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+            dst[(dst_offset>>2)+dy*(dstStep>>2)+dx]= (sx0>=0 && sx0<src_cols && sy0>=0 && sy0<src_rows) ? src[(src_offset>>2)+sy0*(srcStep>>2)+sx0] : (uchar4)0;
+    }
+}
+
+__kernel void warpAffineLinear_C4_D0(__global uchar4 const * restrict src, __global uchar4 * dst, int src_cols, int src_rows,
+                                     int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                     int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/INTER_TAB_SIZE/2;
+
+        src_offset = (src_offset>>2);
+        srcStep = (srcStep>>2);
+
+        int tmp = (dx << AB_BITS);
+        int X0 = rint(M[0] * tmp);
+        int Y0 = rint(M[3] * tmp);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        X0 = X0 >> (AB_BITS - INTER_BITS);
+        Y0 = Y0 >> (AB_BITS - INTER_BITS);
+
+        short sx0 = (short)(X0 >> INTER_BITS);
+        short sy0 = (short)(Y0 >> INTER_BITS);
+        short ax0 = (short)(X0 & (INTER_TAB_SIZE-1));
+        short ay0 = (short)(Y0 & (INTER_TAB_SIZE-1));
+
+        int4 v0, v1, v2, v3;
+
+        v0 = (sx0 >= 0 && sx0 < src_cols && sy0 >= 0 && sy0 < src_rows) ? convert_int4(src[src_offset+sy0 * srcStep + sx0]) : 0;
+        v1 = (sx0+1 >= 0 && sx0+1 < src_cols && sy0 >= 0 && sy0 < src_rows) ? convert_int4(src[src_offset+sy0 * srcStep + sx0+1]) : 0;
+        v2 = (sx0 >= 0 && sx0 < src_cols && sy0+1 >= 0 && sy0+1 < src_rows) ? convert_int4(src[src_offset+(sy0+1) * srcStep + sx0]) : 0;
+        v3 = (sx0+1 >= 0 && sx0+1 < src_cols && sy0+1 >= 0 && sy0+1 < src_rows) ? convert_int4(src[src_offset+(sy0+1) * srcStep + sx0+1]) : 0;
+
+        int itab0, itab1, itab2, itab3;
+        float taby, tabx;
+        taby = 1.f/INTER_TAB_SIZE*ay0;
+        tabx = 1.f/INTER_TAB_SIZE*ax0;
+
+        itab0 = convert_short_sat(rint( (1.0f-taby)*(1.0f-tabx) * INTER_REMAP_COEF_SCALE ));
+        itab1 = convert_short_sat(rint( (1.0f-taby)*tabx * INTER_REMAP_COEF_SCALE ));
+        itab2 = convert_short_sat(rint( taby*(1.0f-tabx) * INTER_REMAP_COEF_SCALE ));
+        itab3 = convert_short_sat(rint( taby*tabx * INTER_REMAP_COEF_SCALE ));
+
+        int4 val;
+        val = v0 * itab0 +  v1 * itab1 + v2 * itab2 + v3 * itab3;
+
+        if(dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+            dst[(dst_offset>>2)+dy*(dstStep>>2)+dx] =  convert_uchar4_sat ( (val + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS ) ;
+    }
+}
+
+__kernel void warpAffineCubic_C4_D0(__global uchar4 const * restrict src, __global uchar4 * dst, int src_cols, int src_rows,
+                                    int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                    int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = ((AB_SCALE>>INTER_BITS)>>1);
+
+        src_offset = (src_offset>>2);
+        srcStep = (srcStep>>2);
+        dst_offset = (dst_offset>>2);
+        dstStep = (dstStep>>2);
+
+        int tmp = (dx << AB_BITS);
+        int X0 = rint(M[0] * tmp);
+        int Y0 = rint(M[3] * tmp);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        X0 = X0 >> (AB_BITS - INTER_BITS);
+        Y0 = Y0 >> (AB_BITS - INTER_BITS);
+
+        int sx = (short)(X0 >> INTER_BITS) - 1;
+        int sy = (short)(Y0 >> INTER_BITS) - 1;
+        int ay = (short)(Y0 & (INTER_TAB_SIZE-1));
+        int ax = (short)(X0 & (INTER_TAB_SIZE-1));
+
+        uchar4 v[16];
+        int i,j;
+#pragma unroll 4
+        for(i=0; i<4; i++)
+            for(j=0; j<4; j++)
+            {
+                v[i*4+j] = (sx+j >= 0 && sx+j < src_cols && sy+i >= 0 && sy+i < src_rows) ? (src[src_offset+(sy+i) * srcStep + (sx+j)])  : (uchar4)0;
+            }
+        int itab[16];
+        float tab1y[4], tab1x[4];
+        float axx, ayy;
+
+        ayy = INTER_SCALE * ay;
+        axx = INTER_SCALE * ax;
+        interpolateCubic(ayy, tab1y);
+        interpolateCubic(axx, tab1x);
+        int isum = 0;
+
+#pragma unroll 16
+        for( i=0; i<16; i++ )
+        {
+            float tmp;
+            tmp = tab1y[(i>>2)] * tab1x[(i&3)] * INTER_REMAP_COEF_SCALE;
+            itab[i] = rint(tmp);
+            isum += itab[i];
+        }
+
+        if( isum != INTER_REMAP_COEF_SCALE )
+        {
+            int k1, k2;
+            int diff = isum - INTER_REMAP_COEF_SCALE;
+            int Mk1=2, Mk2=2, mk1=2, mk2=2;
+
+            for( k1 = 2; k1 < 4; k1++ )
+                for( k2 = 2; k2 < 4; k2++ )
+                {
+
+                    if( itab[(k1<<2)+k2] < itab[(mk1<<2)+mk2] )
+                        mk1 = k1, mk2 = k2;
+                    else if( itab[(k1<<2)+k2] > itab[(Mk1<<2)+Mk2] )
+                        Mk1 = k1, Mk2 = k2;
+                }
+
+            diff<0 ? (itab[(Mk1<<2)+Mk2]=(short)(itab[(Mk1<<2)+Mk2]-diff)) : (itab[(mk1<<2)+mk2]=(short)(itab[(mk1<<2)+mk2]-diff));
+        }
+
+        if( dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+        {
+            int4 sum=0;
+            for ( i =0; i<16; i++ )
+            {
+                sum += convert_int4(v[i]) * itab[i];
+            }
+            dst[dst_offset+dy*dstStep+dx] = convert_uchar4_sat( (sum + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS ) ;
+        }
+    }
+}
+
+
+/**********************************************32FC1********************************************
+***********************************************************************************************/
+
+__kernel void warpAffineNN_C1_D5(__global float * src, __global float * dst, int src_cols, int src_rows,
+                                 int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                 int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/2;
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+
+        short sx0 = (short)(X0 >> AB_BITS);
+        short sy0 = (short)(Y0 >> AB_BITS);
+
+        if(dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+            dst[(dst_offset>>2)+dy*dstStep+dx]= (sx0>=0 && sx0<src_cols && sy0>=0 && sy0<src_rows) ? src[(src_offset>>2)+sy0*srcStep+sx0] : 0;
+    }
+}
+
+__kernel void warpAffineLinear_C1_D5(__global float * src, __global float * dst, int src_cols, int src_rows,
+                                     int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                     int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/INTER_TAB_SIZE/2;
+
+        src_offset = (src_offset>>2);
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        X0 = X0 >> (AB_BITS - INTER_BITS);
+        Y0 = Y0 >> (AB_BITS - INTER_BITS);
+
+        short sx0 = (short)(X0 >> INTER_BITS);
+        short sy0 = (short)(Y0 >> INTER_BITS);
+        short ax0 = (short)(X0 & (INTER_TAB_SIZE-1));
+        short ay0 = (short)(Y0 & (INTER_TAB_SIZE-1));
+
+        float v0, v1, v2, v3;
+
+        v0 = (sx0 >= 0 && sx0 < src_cols && sy0 >= 0 && sy0 < src_rows) ? src[src_offset+sy0 * srcStep + sx0] : 0;
+        v1 = (sx0+1 >= 0 && sx0+1 < src_cols && sy0 >= 0 && sy0 < src_rows) ? src[src_offset+sy0 * srcStep + sx0+1] : 0;
+        v2 = (sx0 >= 0 && sx0 < src_cols && sy0+1 >= 0 && sy0+1 < src_rows) ? src[src_offset+(sy0+1) * srcStep + sx0] : 0;
+        v3 = (sx0+1 >= 0 && sx0+1 < src_cols && sy0+1 >= 0 && sy0+1 < src_rows) ? src[src_offset+(sy0+1) * srcStep + sx0+1] : 0;
+
+        float tab[4];
+        float taby[2], tabx[2];
+        taby[0] = 1.0f - 1.f/INTER_TAB_SIZE*ay0;
+        taby[1] = 1.f/INTER_TAB_SIZE*ay0;
+        tabx[0] = 1.0f - 1.f/INTER_TAB_SIZE*ax0;
+        tabx[1] = 1.f/INTER_TAB_SIZE*ax0;
+
+        tab[0] = taby[0] * tabx[0];
+        tab[1] = taby[0] * tabx[1];
+        tab[2] = taby[1] * tabx[0];
+        tab[3] = taby[1] * tabx[1];
+
+        float sum = 0;
+        sum += v0 * tab[0] +  v1 * tab[1] +  v2 * tab[2] +  v3 * tab[3];
+        if(dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+            dst[(dst_offset>>2)+dy*dstStep+dx] = sum;
+    }
+}
+
+__kernel void warpAffineCubic_C1_D5(__global float * src, __global float * dst, int src_cols, int src_rows,
+                                    int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                    int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/INTER_TAB_SIZE/2;
+
+        src_offset = (src_offset>>2);
+        dst_offset = (dst_offset>>2);
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        X0 = X0 >> (AB_BITS - INTER_BITS);
+        Y0 = Y0 >> (AB_BITS - INTER_BITS);
+
+        short sx = (short)(X0 >> INTER_BITS) - 1;
+        short sy = (short)(Y0 >> INTER_BITS) - 1;
+        short ay = (short)(Y0 & (INTER_TAB_SIZE-1));
+        short ax = (short)(X0 & (INTER_TAB_SIZE-1));
+
+        float v[16];
+        int i;
+
+        for(i=0; i<16;  i++)
+            v[i] = (sx+(i&3) >= 0 && sx+(i&3) < src_cols && sy+(i>>2) >= 0 && sy+(i>>2) < src_rows) ? src[src_offset+(sy+(i>>2)) * srcStep + (sx+(i&3))] : 0;
+
+        float tab[16];
+        float tab1y[4], tab1x[4];
+        float axx, ayy;
+
+        ayy = 1.f/INTER_TAB_SIZE * ay;
+        axx = 1.f/INTER_TAB_SIZE * ax;
+        interpolateCubic(ayy, tab1y);
+        interpolateCubic(axx, tab1x);
+
+#pragma unroll 4
+        for( i=0; i<16; i++ )
+        {
+            tab[i] = tab1y[(i>>2)] * tab1x[(i&3)];
+        }
+
+        if( dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+        {
+            float sum = 0;
+#pragma unroll 4
+            for ( i =0; i<16; i++ )
+            {
+                sum += v[i] * tab[i];
+            }
+            dst[dst_offset+dy*dstStep+dx] = sum;
+
+        }
+    }
+}
+
+
+/**********************************************32FC4********************************************
+***********************************************************************************************/
+
+__kernel void warpAffineNN_C4_D5(__global float4 * src, __global float4 * dst, int src_cols, int src_rows,
+                                 int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                 int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/2;
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+
+        short sx0 = (short)(X0 >> AB_BITS);
+        short sy0 = (short)(Y0 >> AB_BITS);
+
+        if(dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+            dst[(dst_offset>>4)+dy*(dstStep>>2)+dx]= (sx0>=0 && sx0<src_cols && sy0>=0 && sy0<src_rows) ? src[(src_offset>>4)+sy0*(srcStep>>2)+sx0] : (float4)0;
+    }
+}
+
+__kernel void warpAffineLinear_C4_D5(__global float4 * src, __global float4 * dst, int src_cols, int src_rows,
+                                     int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                     int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/INTER_TAB_SIZE/2;
+
+        src_offset = (src_offset>>4);
+        dst_offset = (dst_offset>>4);
+        srcStep = (srcStep>>2);
+        dstStep = (dstStep>>2);
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        X0 = X0 >> (AB_BITS - INTER_BITS);
+        Y0 = Y0 >> (AB_BITS - INTER_BITS);
+
+        short sx0 = (short)(X0 >> INTER_BITS);
+        short sy0 = (short)(Y0 >> INTER_BITS);
+        short ax0 = (short)(X0 & (INTER_TAB_SIZE-1));
+        short ay0 = (short)(Y0 & (INTER_TAB_SIZE-1));
+
+        float4 v0, v1, v2, v3;
+
+        v0 = (sx0 >= 0 && sx0 < src_cols && sy0 >= 0 && sy0 < src_rows) ? src[src_offset+sy0 * srcStep + sx0] : (float4)0;
+        v1 = (sx0+1 >= 0 && sx0+1 < src_cols && sy0 >= 0 && sy0 < src_rows) ? src[src_offset+sy0 * srcStep + sx0+1] : (float4)0;
+        v2 = (sx0 >= 0 && sx0 < src_cols && sy0+1 >= 0 && sy0+1 < src_rows) ? src[src_offset+(sy0+1) * srcStep + sx0] : (float4)0;
+        v3 = (sx0+1 >= 0 && sx0+1 < src_cols && sy0+1 >= 0 && sy0+1 < src_rows) ? src[src_offset+(sy0+1) * srcStep + sx0+1] : (float4)0;
+
+        float tab[4];
+        float taby[2], tabx[2];
+        taby[0] = 1.0f - 1.f/INTER_TAB_SIZE*ay0;
+        taby[1] = 1.f/INTER_TAB_SIZE*ay0;
+        tabx[0] = 1.0f - 1.f/INTER_TAB_SIZE*ax0;
+        tabx[1] = 1.f/INTER_TAB_SIZE*ax0;
+
+        tab[0] = taby[0] * tabx[0];
+        tab[1] = taby[0] * tabx[1];
+        tab[2] = taby[1] * tabx[0];
+        tab[3] = taby[1] * tabx[1];
+
+        float4 sum = 0;
+        sum += v0 * tab[0] +  v1 * tab[1] +  v2 * tab[2] +  v3 * tab[3];
+        if(dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+            dst[dst_offset+dy*dstStep+dx] = sum;
+    }
+}
+
+__kernel void warpAffineCubic_C4_D5(__global float4 * src, __global float4 * dst, int src_cols, int src_rows,
+                                    int dst_cols, int dst_rows, int srcStep, int dstStep,
+                                    int src_offset, int dst_offset,  __constant F * M, int threadCols )
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if( dx < threadCols && dy < dst_rows)
+    {
+        int round_delta = AB_SCALE/INTER_TAB_SIZE/2;
+
+        src_offset = (src_offset>>4);
+        dst_offset = (dst_offset>>4);
+        srcStep = (srcStep>>2);
+        dstStep = (dstStep>>2);
+
+        int X0 = rint(M[0] * dx * AB_SCALE);
+        int Y0 = rint(M[3] * dx * AB_SCALE);
+        X0 += rint((M[1]*dy + M[2]) * AB_SCALE) + round_delta;
+        Y0 += rint((M[4]*dy + M[5]) * AB_SCALE) + round_delta;
+        X0 = X0 >> (AB_BITS - INTER_BITS);
+        Y0 = Y0 >> (AB_BITS - INTER_BITS);
+
+        short sx = (short)(X0 >> INTER_BITS) - 1;
+        short sy = (short)(Y0 >> INTER_BITS) - 1;
+        short ay = (short)(Y0 & (INTER_TAB_SIZE-1));
+        short ax = (short)(X0 & (INTER_TAB_SIZE-1));
+
+        float4 v[16];
+        int i;
+
+        for(i=0; i<16;  i++)
+            v[i] = (sx+(i&3) >= 0 && sx+(i&3) < src_cols && sy+(i>>2) >= 0 && sy+(i>>2) < src_rows) ? src[src_offset+(sy+(i>>2)) * srcStep + (sx+(i&3))] : (float4)0;
+
+        float tab[16];
+        float tab1y[4], tab1x[4];
+        float axx, ayy;
+
+        ayy = 1.f/INTER_TAB_SIZE * ay;
+        axx = 1.f/INTER_TAB_SIZE * ax;
+        interpolateCubic(ayy, tab1y);
+        interpolateCubic(axx, tab1x);
+
+#pragma unroll 4
+        for( i=0; i<16; i++ )
+        {
+            tab[i] = tab1y[(i>>2)] * tab1x[(i&3)];
+        }
+
+        if( dx >= 0 && dx < dst_cols && dy >= 0 && dy < dst_rows)
+        {
+            float4 sum = 0;
+#pragma unroll 4
+            for ( i =0; i<16; i++ )
+            {
+                sum += v[i] * tab[i];
+            }
+            dst[dst_offset+dy*dstStep+dx] = sum;
+
+        }
+    }
+}
diff --git a/modules/imgproc/src/opencl/warp_perspective.cl b/modules/imgproc/src/opencl/warp_perspective.cl
new file mode 100644
index 0000000000..211433e709
--- /dev/null
+++ b/modules/imgproc/src/opencl/warp_perspective.cl
@@ -0,0 +1,223 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                           License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
+// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// @Authors
+//    Zhang Ying, zhangying913@gmail.com
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors as is and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#ifdef DOUBLE_SUPPORT
+#ifdef cl_amd_fp64
+#pragma OPENCL EXTENSION cl_amd_fp64:enable
+#elif defined (cl_khr_fp64)
+#pragma OPENCL EXTENSION cl_khr_fp64:enable
+#endif
+#define CT double
+#else
+#define CT float
+#endif
+
+#define INTER_BITS 5
+#define INTER_TAB_SIZE (1 << INTER_BITS)
+#define INTER_SCALE 1.f / INTER_TAB_SIZE
+#define AB_BITS max(10, (int)INTER_BITS)
+#define AB_SCALE (1 << AB_BITS)
+#define INTER_REMAP_COEF_BITS 15
+#define INTER_REMAP_COEF_SCALE (1 << INTER_REMAP_COEF_BITS)
+
+#define noconvert
+
+#ifdef INTER_NEAREST
+
+__kernel void warpPerspective(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
+                              __global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
+                              __constant CT * M, T scalar)
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if (dx < dst_cols && dy < dst_rows)
+    {
+        CT X0 = M[0] * dx + M[1] * dy + M[2];
+        CT Y0 = M[3] * dx + M[4] * dy + M[5];
+        CT W = M[6] * dx + M[7] * dy + M[8];
+        W = W != 0.0f ? 1.f / W : 0.0f;
+        short sx = convert_short_sat_rte(X0*W);
+        short sy = convert_short_sat_rte(Y0*W);
+
+        int dst_index = mad24(dy, dst_step, dx * (int)sizeof(T) + dst_offset);
+        __global T * dst = (__global T *)(dstptr + dst_index);
+
+        if (sx >= 0 && sx < src_cols && sy >= 0 && sy < src_rows)
+        {
+            int src_index = mad24(sy, src_step, sx * (int)sizeof(T) + src_offset);
+            __global const T * src = (__global const T *)(srcptr + src_index);
+            dst[0] = src[0];
+        }
+        else
+            dst[0] = scalar;
+    }
+}
+
+#elif defined INTER_LINEAR
+
+__kernel void warpPerspective(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
+                              __global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
+                              __constant CT * M, WT scalar)
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if (dx < dst_cols && dy < dst_rows)
+    {
+        CT X0 = M[0] * dx + M[1] * dy + M[2];
+        CT Y0 = M[3] * dx + M[4] * dy + M[5];
+        CT W = M[6] * dx + M[7] * dy + M[8];
+        W = W != 0.0f ? INTER_TAB_SIZE / W : 0.0f;
+        int X = rint(X0 * W), Y = rint(Y0 * W);
+
+        short sx = convert_short_sat(X >> INTER_BITS);
+        short sy = convert_short_sat(Y >> INTER_BITS);
+        short ay = (short)(Y & (INTER_TAB_SIZE - 1));
+        short ax = (short)(X & (INTER_TAB_SIZE - 1));
+
+        WT v0 = (sx >= 0 && sx < src_cols && sy >= 0 && sy < src_rows) ?
+            convertToWT(*(__global const T *)(srcptr + mad24(sy, src_step, src_offset + sx * (int)sizeof(T)))) : scalar;
+        WT v1 = (sx+1 >= 0 && sx+1 < src_cols && sy >= 0 && sy < src_rows) ?
+            convertToWT(*(__global const T *)(srcptr + mad24(sy, src_step, src_offset + (sx+1) * (int)sizeof(T)))) : scalar;
+        WT v2 = (sx >= 0 && sx < src_cols && sy+1 >= 0 && sy+1 < src_rows) ?
+            convertToWT(*(__global const T *)(srcptr + mad24(sy+1, src_step, src_offset + sx * (int)sizeof(T)))) : scalar;
+        WT v3 = (sx+1 >= 0 && sx+1 < src_cols && sy+1 >= 0 && sy+1 < src_rows) ?
+            convertToWT(*(__global const T *)(srcptr + mad24(sy+1, src_step, src_offset + (sx+1) * (int)sizeof(T)))) : scalar;
+
+        float taby = 1.f/INTER_TAB_SIZE*ay;
+        float tabx = 1.f/INTER_TAB_SIZE*ax;
+
+        int dst_index = mad24(dy, dst_step, dst_offset + dx * (int)sizeof(T));
+        __global T * dst = (__global T *)(dstptr + dst_index);
+
+#if depth <= 4
+        int itab0 = convert_short_sat_rte( (1.0f-taby)*(1.0f-tabx) * INTER_REMAP_COEF_SCALE );
+        int itab1 = convert_short_sat_rte( (1.0f-taby)*tabx * INTER_REMAP_COEF_SCALE );
+        int itab2 = convert_short_sat_rte( taby*(1.0f-tabx) * INTER_REMAP_COEF_SCALE );
+        int itab3 = convert_short_sat_rte( taby*tabx * INTER_REMAP_COEF_SCALE );
+
+        WT val = v0 * itab0 +  v1 * itab1 + v2 * itab2 + v3 * itab3;
+        dst[0] = convertToT((val + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS);
+#else
+        float tabx2 = 1.0f - tabx, taby2 = 1.0f - taby;
+        WT val = v0 * tabx2 * taby2 +  v1 * tabx * taby2 + v2 * tabx2 * taby + v3 * tabx * taby;
+        dst[0] = convertToT(val);
+#endif
+    }
+}
+
+#elif defined INTER_CUBIC
+
+inline void interpolateCubic( float x, float* coeffs )
+{
+    const float A = -0.75f;
+
+    coeffs[0] = ((A*(x + 1.f) - 5.0f*A)*(x + 1.f) + 8.0f*A)*(x + 1.f) - 4.0f*A;
+    coeffs[1] = ((A + 2.f)*x - (A + 3.f))*x*x + 1.f;
+    coeffs[2] = ((A + 2.f)*(1.f - x) - (A + 3.f))*(1.f - x)*(1.f - x) + 1.f;
+    coeffs[3] = 1.f - coeffs[0] - coeffs[1] - coeffs[2];
+}
+
+__kernel void warpPerspective(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
+                              __global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
+                              __constant CT * M, WT scalar)
+{
+    int dx = get_global_id(0);
+    int dy = get_global_id(1);
+
+    if (dx < dst_cols && dy < dst_rows)
+    {
+        CT X0 = M[0] * dx + M[1] * dy + M[2];
+        CT Y0 = M[3] * dx + M[4] * dy + M[5];
+        CT W = M[6] * dx + M[7] * dy + M[8];
+        W = W != 0.0f ? INTER_TAB_SIZE / W : 0.0f;
+        int X = rint(X0 * W), Y = rint(Y0 * W);
+
+        short sx = convert_short_sat(X >> INTER_BITS) - 1;
+        short sy = convert_short_sat(Y >> INTER_BITS) - 1;
+        short ay = (short)(Y & (INTER_TAB_SIZE-1));
+        short ax = (short)(X & (INTER_TAB_SIZE-1));
+
+        WT v[16];
+        #pragma unroll
+        for (int y = 0; y < 4; y++)
+            #pragma unroll
+            for (int x = 0; x < 4; x++)
+                v[mad24(y, 4, x)] = (sx+x >= 0 && sx+x < src_cols && sy+y >= 0 && sy+y < src_rows) ?
+                    convertToWT(*(__global const T *)(srcptr + mad24(sy+y, src_step, src_offset + (sx+x) * (int)sizeof(T)))) : scalar;
+
+        float tab1y[4], tab1x[4];
+
+        float ayy = INTER_SCALE * ay;
+        float axx = INTER_SCALE * ax;
+        interpolateCubic(ayy, tab1y);
+        interpolateCubic(axx, tab1x);
+
+        int dst_index = mad24(dy, dst_step, dst_offset + dx * (int)sizeof(T));
+        __global T * dst = (__global T *)(dstptr + dst_index);
+
+        WT sum = (WT)(0);
+#if depth <= 4
+        int itab[16];
+
+        #pragma unroll
+        for (int i = 0; i < 16; i++)
+            itab[i] = rint(tab1y[(i>>2)] * tab1x[(i&3)] * INTER_REMAP_COEF_SCALE);
+
+        #pragma unroll
+        for (int i = 0; i < 16; i++)
+            sum += v[i] * itab[i];
+        dst[0] = convertToT( (sum + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS );
+#else
+        #pragma unroll
+        for (int i = 0; i < 16; i++)
+            sum += v[i] * tab1y[(i>>2)] * tab1x[(i&3)];
+        dst[0] = convertToT( sum );
+#endif
+    }
+}
+
+#endif
diff --git a/modules/imgproc/test/ocl/test_warp.cpp b/modules/imgproc/test/ocl/test_warp.cpp
index 6e549a4eca..fece5e7e45 100644
--- a/modules/imgproc/test/ocl/test_warp.cpp
+++ b/modules/imgproc/test/ocl/test_warp.cpp
@@ -61,7 +61,99 @@ namespace cvtest {
 namespace ocl {
 
 /////////////////////////////////////////////////////////////////////////////////////////////////
-// resize
+// warpAffine  & warpPerspective
+
+PARAM_TEST_CASE(WarpTestBase, MatType, Interpolation, bool, bool)
+{
+    int type, interpolation;
+    Size dsize;
+    bool useRoi, mapInverse;
+
+    TEST_DECLARE_INPUT_PARATEMER(src)
+    TEST_DECLARE_OUTPUT_PARATEMER(dst)
+
+    virtual void SetUp()
+    {
+        type = GET_PARAM(0);
+        interpolation = GET_PARAM(1);
+        mapInverse = GET_PARAM(2);
+        useRoi = GET_PARAM(3);
+
+        if (mapInverse)
+            interpolation |= WARP_INVERSE_MAP;
+    }
+
+    void random_roi()
+    {
+        dsize = randomSize(1, MAX_VALUE);
+
+        Size roiSize = randomSize(1, MAX_VALUE);
+        Border srcBorder = randomBorder(0, useRoi ? MAX_VALUE : 0);
+        randomSubMat(src, src_roi, roiSize, srcBorder, type, -MAX_VALUE, MAX_VALUE);
+
+        Border dstBorder = randomBorder(0, useRoi ? MAX_VALUE : 0);
+        randomSubMat(dst, dst_roi, dsize, dstBorder, type, -MAX_VALUE, MAX_VALUE);
+
+        UMAT_UPLOAD_INPUT_PARAMETER(src)
+        UMAT_UPLOAD_OUTPUT_PARAMETER(dst)
+    }
+
+    void Near(double threshold = 0.0)
+    {
+        EXPECT_MAT_NEAR(dst, udst, threshold);
+        EXPECT_MAT_NEAR(dst_roi, udst_roi, threshold);
+    }
+};
+
+/////warpAffine
+
+typedef WarpTestBase WarpAffine;
+
+OCL_TEST_P(WarpAffine, Mat)
+{
+    for (int j = 0; j < test_loop_times; j++)
+    {
+        random_roi();
+
+        Mat M = getRotationMatrix2D(Point2f(src_roi.cols / 2.0f, src_roi.rows / 2.0f),
+            rng.uniform(-180.f, 180.f), rng.uniform(0.4f, 2.0f));
+
+        OCL_OFF(cv::warpAffine(src_roi, dst_roi, M, dsize, interpolation));
+        OCL_ON(cv::warpAffine(usrc_roi, udst_roi, M, dsize, interpolation));
+
+        Near(1.0);
+    }
+}
+
+//// warpPerspective
+
+typedef WarpTestBase WarpPerspective;
+
+OCL_TEST_P(WarpPerspective, Mat)
+{
+    for (int j = 0; j < test_loop_times; j++)
+    {
+        random_roi();
+
+        float cols = static_cast<float>(src_roi.cols), rows = static_cast<float>(src_roi.rows);
+        float cols2 = cols / 2.0f, rows2 = rows / 2.0f;
+        Point2f sp[] = { Point2f(0.0f, 0.0f), Point2f(cols, 0.0f), Point2f(0.0f, rows), Point2f(cols, rows) };
+        Point2f dp[] = { Point2f(rng.uniform(0.0f, cols2), rng.uniform(0.0f, rows2)),
+            Point2f(rng.uniform(cols2, cols), rng.uniform(0.0f, rows2)),
+            Point2f(rng.uniform(0.0f, cols2), rng.uniform(rows2, rows)),
+            Point2f(rng.uniform(cols2, cols), rng.uniform(rows2, rows)) };
+        Mat M = getPerspectiveTransform(sp, dp);
+
+        OCL_OFF(cv::warpPerspective(src_roi, dst_roi, M, dsize, interpolation));
+        OCL_ON(cv::warpPerspective(usrc_roi, udst_roi, M, dsize, interpolation));
+
+        Near(1.0);
+    }
+}
+
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+//// resize
 
 PARAM_TEST_CASE(Resize, MatType, double, double, Interpolation, bool)
 {
@@ -127,10 +219,22 @@ OCL_TEST_P(Resize, Mat)
 
 /////////////////////////////////////////////////////////////////////////////////////
 
-OCL_INSTANTIATE_TEST_CASE_P(ImgprocWarpResize, Resize, Combine(
-                            Values((MatType)CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4),
-                            Values(0.7, 0.4, 2.0),
-                            Values(0.3, 0.6, 2.0),
+OCL_INSTANTIATE_TEST_CASE_P(ImgprocWarp, WarpAffine, Combine(
+                            Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
+                            Values((Interpolation)INTER_NEAREST, (Interpolation)INTER_LINEAR, (Interpolation)INTER_CUBIC),
+                            Bool(),
+                            Bool()));
+
+OCL_INSTANTIATE_TEST_CASE_P(ImgprocWarp, WarpPerspective, Combine(
+                            Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
+                            Values((Interpolation)INTER_NEAREST, (Interpolation)INTER_LINEAR, (Interpolation)INTER_CUBIC),
+                            Bool(),
+                            Bool()));
+
+OCL_INSTANTIATE_TEST_CASE_P(ImgprocWarp, Resize, Combine(
+                            Values(CV_8UC1, CV_8UC4, CV_16UC2, CV_32FC1, CV_32FC4),
+                            Values(0.5, 1.5, 2.0),
+                            Values(0.5, 1.5, 2.0),
                             Values((Interpolation)INTER_NEAREST, (Interpolation)INTER_LINEAR),
                             Bool()));