feat: medianBlur & bilateralFilter

3rdparty/ndsrvp/include/imgproc.hpp

@@ -101,6 +101,24 @@ int filterFree(cvhalFilter2D *context);
 #undef cv_hal_filterFree
 #define cv_hal_filterFree (cv::ndsrvp::filterFree)
 
+// ################ medianBlur ################
+
+int medianBlur(const uchar* src_data, size_t src_step,
+    uchar* dst_data, size_t dst_step,
+    int width, int height, int depth, int cn, int ksize);
+
+#undef cv_hal_medianBlur
+#define cv_hal_medianBlur (cv::ndsrvp::medianBlur)
+
+// ################ bilateralFilter ################
+
+int bilateralFilter(const uchar* src_data, size_t src_step,
+    uchar* dst_data, size_t dst_step, int width, int height, int depth,
+    int cn, int d, double sigma_color, double sigma_space, int border_type);
+
+#undef cv_hal_bilateralFilter
+#define cv_hal_bilateralFilter (cv::ndsrvp::bilateralFilter)
+
 } // namespace ndsrvp
 
 } // namespace cv

3rdparty/ndsrvp/src/bilateralFilter.cpp

@@ -0,0 +1,270 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#include "ndsrvp_hal.hpp"
+#include "opencv2/imgproc/hal/interface.h"
+#include "cvutils.hpp"
+
+namespace cv {
+
+namespace ndsrvp {
+
+static void bilateralFilterProcess(uchar* dst_data, size_t dst_step, uchar* pad_data, size_t pad_step,
+    int width, int height, int cn, int radius, int maxk,
+    int* space_ofs, float *space_weight, float *color_weight)
+{
+    int i, j, k;
+
+    for( i = 0; i < height; i++ )
+    {
+        const uchar* sptr = pad_data + (i + radius) * pad_step + radius * cn;
+        uchar* dptr = dst_data + i * dst_step;
+
+        if( cn == 1 )
+        {
+            std::vector<float> buf(width + width, 0.0);
+            float *sum = &buf[0];
+            float *wsum = sum + width;
+            k = 0;
+            for(; k <= maxk-4; k+=4)
+            {
+                const uchar* ksptr0 = sptr + space_ofs[k];
+                const uchar* ksptr1 = sptr + space_ofs[k+1];
+                const uchar* ksptr2 = sptr + space_ofs[k+2];
+                const uchar* ksptr3 = sptr + space_ofs[k+3];
+                j = 0;
+                for (; j < width; j++)
+                {
+                    int rval = sptr[j];
+
+                    int val = ksptr0[j];
+                    float w = space_weight[k] * color_weight[std::abs(val - rval)];
+                    wsum[j] += w;
+                    sum[j] += val * w;
+
+                    val = ksptr1[j];
+                    w = space_weight[k+1] * color_weight[std::abs(val - rval)];
+                    wsum[j] += w;
+                    sum[j] += val * w;
+
+                    val = ksptr2[j];
+                    w = space_weight[k+2] * color_weight[std::abs(val - rval)];
+                    wsum[j] += w;
+                    sum[j] += val * w;
+
+                    val = ksptr3[j];
+                    w = space_weight[k+3] * color_weight[std::abs(val - rval)];
+                    wsum[j] += w;
+                    sum[j] += val * w;
+                }
+            }
+            for(; k < maxk; k++)
+            {
+                const uchar* ksptr = sptr + space_ofs[k];
+                j = 0;
+                for (; j < width; j++)
+                {
+                    int val = ksptr[j];
+                    float w = space_weight[k] * color_weight[std::abs(val - sptr[j])];
+                    wsum[j] += w;
+                    sum[j] += val * w;
+                }
+            }
+            j = 0;
+            for (; j < width; j++)
+            {
+                // overflow is not possible here => there is no need to use cv::saturate_cast
+                ndsrvp_assert(fabs(wsum[j]) > 0);
+                dptr[j] = (uchar)(sum[j] / wsum[j] + 0.5);
+            }
+        }
+        else
+        {
+            ndsrvp_assert( cn == 3 );
+            std::vector<float> buf(width * 3 + width);
+            float *sum_b = &buf[0];
+            float *sum_g = sum_b + width;
+            float *sum_r = sum_g + width;
+            float *wsum = sum_r + width;
+            k = 0;
+            for(; k <= maxk-4; k+=4)
+            {
+                const uchar* ksptr0 = sptr + space_ofs[k];
+                const uchar* ksptr1 = sptr + space_ofs[k+1];
+                const uchar* ksptr2 = sptr + space_ofs[k+2];
+                const uchar* ksptr3 = sptr + space_ofs[k+3];
+                const uchar* rsptr = sptr;
+                j = 0;
+                for(; j < width; j++, rsptr += 3, ksptr0 += 3, ksptr1 += 3, ksptr2 += 3, ksptr3 += 3)
+                {
+                    int rb = rsptr[0], rg = rsptr[1], rr = rsptr[2];
+
+                    int b = ksptr0[0], g = ksptr0[1], r = ksptr0[2];
+                    float w = space_weight[k] * color_weight[std::abs(b - rb) + std::abs(g - rg) + std::abs(r - rr)];
+                    wsum[j] += w;
+                    sum_b[j] += b * w; sum_g[j] += g * w; sum_r[j] += r * w;
+
+                    b = ksptr1[0]; g = ksptr1[1]; r = ksptr1[2];
+                    w = space_weight[k+1] * color_weight[std::abs(b - rb) + std::abs(g - rg) + std::abs(r - rr)];
+                    wsum[j] += w;
+                    sum_b[j] += b * w; sum_g[j] += g * w; sum_r[j] += r * w;
+
+                    b = ksptr2[0]; g = ksptr2[1]; r = ksptr2[2];
+                    w = space_weight[k+2] * color_weight[std::abs(b - rb) + std::abs(g - rg) + std::abs(r - rr)];
+                    wsum[j] += w;
+                    sum_b[j] += b * w; sum_g[j] += g * w; sum_r[j] += r * w;
+
+                    b = ksptr3[0]; g = ksptr3[1]; r = ksptr3[2];
+                    w = space_weight[k+3] * color_weight[std::abs(b - rb) + std::abs(g - rg) + std::abs(r - rr)];
+                    wsum[j] += w;
+                    sum_b[j] += b * w; sum_g[j] += g * w; sum_r[j] += r * w;
+                }
+            }
+            for(; k < maxk; k++)
+            {
+                const uchar* ksptr = sptr + space_ofs[k];
+                const uchar* rsptr = sptr;
+                j = 0;
+                for(; j < width; j++, ksptr += 3, rsptr += 3)
+                {
+                    int b = ksptr[0], g = ksptr[1], r = ksptr[2];
+                    float w = space_weight[k] * color_weight[std::abs(b - rsptr[0]) + std::abs(g - rsptr[1]) + std::abs(r - rsptr[2])];
+                    wsum[j] += w;
+                    sum_b[j] += b * w; sum_g[j] += g * w; sum_r[j] += r * w;
+                }
+            }
+            j = 0;
+            for(; j < width; j++)
+            {
+                ndsrvp_assert(fabs(wsum[j]) > 0);
+                wsum[j] = 1.f / wsum[j];
+                *(dptr++) = (uchar)(sum_b[j] * wsum[j] + 0.5);
+                *(dptr++) = (uchar)(sum_g[j] * wsum[j] + 0.5);
+                *(dptr++) = (uchar)(sum_r[j] * wsum[j] + 0.5);
+            }
+        }
+    }
+}
+
+int bilateralFilter(const uchar* src_data, size_t src_step,
+    uchar* dst_data, size_t dst_step, int width, int height, int depth,
+    int cn, int d, double sigma_color, double sigma_space, int border_type)
+{
+    if( depth != CV_8U || !(cn == 1 || cn == 3) || src_data == dst_data)
+        return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+    int i, j, maxk, radius;
+
+    if( sigma_color <= 0 )
+        sigma_color = 1;
+    if( sigma_space <= 0 )
+        sigma_space = 1;
+
+    double gauss_color_coeff = -0.5/(sigma_color * sigma_color);
+    double gauss_space_coeff = -0.5/(sigma_space * sigma_space);
+
+    if( d <= 0 )
+        radius = (int)(sigma_space * 1.5 + 0.5);
+    else
+        radius = d / 2;
+
+    radius = MAX(radius, 1);
+    d = radius * 2 + 1;
+
+    // no enough submatrix info
+    // fetch original image data
+    const uchar *ogn_data = src_data;
+    int ogn_step = src_step;
+
+    // ROI fully used in the computation
+    int cal_width = width + d - 1;
+    int cal_height = height + d - 1;
+    int cal_x = 0 - radius; // negative if left border exceeded
+    int cal_y = 0 - radius; // negative if top border exceeded
+
+    // calculate source border
+    std::vector<uchar> padding;
+    padding.resize(cal_width * cal_height * cn);
+    uchar* pad_data = &padding[0];
+    int pad_step = cal_width * cn;
+
+    uchar* pad_ptr;
+    const uchar* ogn_ptr;
+    std::vector<uchar> vec_zeros(cn, 0);
+    for(i = 0; i < cal_height; i++)
+    {
+        int y = borderInterpolate(i + cal_y, height, border_type);
+        if(y < 0) {
+            memset(pad_data + i * pad_step, 0, cn * cal_width);
+            continue;
+        }
+
+        // left border
+        j = 0;
+        for(; j + cal_x < 0; j++)
+        {
+            int x = borderInterpolate(j + cal_x, width, border_type);
+            if(x < 0) // border constant return value -1
+                ogn_ptr = &vec_zeros[0];
+            else
+                ogn_ptr = ogn_data + y * ogn_step + x * cn;
+            pad_ptr = pad_data + i * pad_step + j * cn;
+            memcpy(pad_ptr, ogn_ptr, cn);
+        }
+
+        // center
+        int rborder = MIN(cal_width, width - cal_x);
+        ogn_ptr = ogn_data + y * ogn_step + (j + cal_x) * cn;
+        pad_ptr = pad_data + i * pad_step + j * cn;
+        memcpy(pad_ptr, ogn_ptr, cn * (rborder - j));
+
+        // right border
+        j = rborder;
+        for(; j < cal_width; j++)
+        {
+            int x = borderInterpolate(j + cal_x, width, border_type);
+            if(x < 0) // border constant return value -1
+                ogn_ptr = &vec_zeros[0];
+            else
+                ogn_ptr = ogn_data + y * ogn_step + x * cn;
+            pad_ptr = pad_data + i * pad_step + j * cn;
+            memcpy(pad_ptr, ogn_ptr, cn);
+        }
+    }
+
+    std::vector<float> _color_weight(cn * 256);
+    std::vector<float> _space_weight(d * d);
+    std::vector<int> _space_ofs(d * d);
+    float* color_weight = &_color_weight[0];
+    float* space_weight = &_space_weight[0];
+    int* space_ofs = &_space_ofs[0];
+
+    // initialize color-related bilateral filter coefficients
+
+    for( i = 0; i < 256 * cn; i++ )
+        color_weight[i] = (float)std::exp(i * i * gauss_color_coeff);
+
+    // initialize space-related bilateral filter coefficients
+    for( i = -radius, maxk = 0; i <= radius; i++ )
+    {
+        j = -radius;
+
+        for( ; j <= radius; j++ )
+        {
+            double r = std::sqrt((double)i * i + (double)j * j);
+            if( r > radius )
+                continue;
+            space_weight[maxk] = (float)std::exp(r * r * gauss_space_coeff);
+            space_ofs[maxk++] = (int)(i * pad_step + j * cn);
+        }
+    }
+
+    bilateralFilterProcess(dst_data, dst_step, pad_data, pad_step, width, height, cn, radius, maxk, space_ofs, space_weight, color_weight);
+
+    return CV_HAL_ERROR_OK;
+}
+
+} // namespace ndsrvp
+
+} // namespace cv

3rdparty/ndsrvp/src/filter.cpp

@@ -132,8 +132,8 @@ int filter(cvhalFilter2D *context,
     // ROI fully used in the computation
     int cal_width = width + ctx->kernel_width - 1;
     int cal_height = height + ctx->kernel_height - 1;
-    int cal_x = offset_x - ctx->anchor_x;
+    int cal_x = offset_x - ctx->anchor_x; // negative if left border exceeded
-    int cal_y = offset_y - ctx->anchor_y;
+    int cal_y = offset_y - ctx->anchor_y; // negative if top border exceeded
 
     // calculate source border
     ctx->padding.resize(cal_width * cal_height * cnes);

3rdparty/ndsrvp/src/medianBlur.cpp

@@ -0,0 +1,300 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#include "ndsrvp_hal.hpp"
+#include "opencv2/imgproc/hal/interface.h"
+#include "cvutils.hpp"
+
+namespace cv {
+
+namespace ndsrvp {
+
+struct operators_minmax_t {
+    inline void vector(uint8x8_t & a, uint8x8_t & b) const {
+        uint8x8_t t = a;
+        a = __nds__v_umin8(a, b);
+        b = __nds__v_umax8(t, b);
+    }
+    inline void scalar(uchar & a, uchar & b) const {
+        uchar t = a;
+        a = __nds__umin8(a, b);
+        b = __nds__umax8(t, b);
+    }
+    inline void vector(int8x8_t & a, int8x8_t & b) const {
+        int8x8_t t = a;
+        a = __nds__v_smin8(a, b);
+        b = __nds__v_smax8(t, b);
+    }
+    inline void scalar(schar & a, schar & b) const {
+        schar t = a;
+        a = __nds__smin8(a, b);
+        b = __nds__smax8(t, b);
+    }
+    inline void vector(uint16x4_t & a, uint16x4_t & b) const {
+        uint16x4_t t = a;
+        a = __nds__v_umin16(a, b);
+        b = __nds__v_umax16(t, b);
+    }
+    inline void scalar(ushort & a, ushort & b) const {
+        ushort t = a;
+        a = __nds__umin16(a, b);
+        b = __nds__umax16(t, b);
+    }
+    inline void vector(int16x4_t & a, int16x4_t & b) const {
+        int16x4_t t = a;
+        a = __nds__v_smin16(a, b);
+        b = __nds__v_smax16(t, b);
+    }
+    inline void scalar(short & a, short & b) const {
+        short t = a;
+        a = __nds__smin16(a, b);
+        b = __nds__smax16(t, b);
+    }
+};
+
+template<typename T, typename WT, typename VT> // type, widen type, vector type
+static void
+medianBlur_SortNet( const uchar* src_data, size_t src_step,
+                    uchar* dst_data, size_t dst_step,
+                    int width, int height, int cn, int ksize )
+{
+    const T* src = (T*)src_data;
+    T* dst = (T*)dst_data;
+    int sstep = (int)(src_step / sizeof(T));
+    int dstep = (int)(dst_step / sizeof(T));
+    int i, j, k;
+    operators_minmax_t op;
+
+    if( ksize == 3 )
+    {
+        if( width == 1 || height == 1 )
+        {
+            int len = width + height - 1;
+            int sdelta = height == 1 ? cn : sstep;
+            int sdelta0 = height == 1 ? 0 : sstep - cn;
+            int ddelta = height == 1 ? cn : dstep;
+
+            for( i = 0; i < len; i++, src += sdelta0, dst += ddelta )
+                for( j = 0; j < cn; j++, src++ )
+                {
+                    T p0 = src[i > 0 ? -sdelta : 0];
+                    T p1 = src[0];
+                    T p2 = src[i < len - 1 ? sdelta : 0];
+
+                    op.scalar(p0, p1); op.scalar(p1, p2); op.scalar(p0, p1);
+                    dst[j] = (T)p1;
+                }
+            return;
+        }
+
+        width *= cn;
+        for( i = 0; i < height; i++, dst += dstep )
+        {
+            const T* row0 = src + std::max(i - 1, 0)*sstep;
+            const T* row1 = src + i*sstep;
+            const T* row2 = src + std::min(i + 1, height-1)*sstep;
+            int limit = cn;
+
+            for(j = 0;; )
+            {
+                for( ; j < limit; j++ )
+                {
+                    int j0 = j >= cn ? j - cn : j;
+                    int j2 = j < width - cn ? j + cn : j;
+                    T p0 = row0[j0], p1 = row0[j], p2 = row0[j2];
+                    T p3 = row1[j0], p4 = row1[j], p5 = row1[j2];
+                    T p6 = row2[j0], p7 = row2[j], p8 = row2[j2];
+
+                    op.scalar(p1, p2); op.scalar(p4, p5); op.scalar(p7, p8); op.scalar(p0, p1);
+                    op.scalar(p3, p4); op.scalar(p6, p7); op.scalar(p1, p2); op.scalar(p4, p5);
+                    op.scalar(p7, p8); op.scalar(p0, p3); op.scalar(p5, p8); op.scalar(p4, p7);
+                    op.scalar(p3, p6); op.scalar(p1, p4); op.scalar(p2, p5); op.scalar(p4, p7);
+                    op.scalar(p4, p2); op.scalar(p6, p4); op.scalar(p4, p2);
+                    dst[j] = (T)p4;
+                }
+
+                if( limit == width )
+                    break;
+
+                int nlanes = 8 / sizeof(T);
+
+                for( ; (cn % nlanes == 0) && (j <= width - nlanes - cn); j += nlanes ) // alignment
+                {
+                    VT p0 = *(VT*)(row0+j-cn), p1 = *(VT*)(row0+j), p2 = *(VT*)(row0+j+cn);
+                    VT p3 = *(VT*)(row1+j-cn), p4 = *(VT*)(row1+j), p5 = *(VT*)(row1+j+cn);
+                    VT p6 = *(VT*)(row2+j-cn), p7 = *(VT*)(row2+j), p8 = *(VT*)(row2+j+cn);
+
+                    op.vector(p1, p2); op.vector(p4, p5); op.vector(p7, p8); op.vector(p0, p1);
+                    op.vector(p3, p4); op.vector(p6, p7); op.vector(p1, p2); op.vector(p4, p5);
+                    op.vector(p7, p8); op.vector(p0, p3); op.vector(p5, p8); op.vector(p4, p7);
+                    op.vector(p3, p6); op.vector(p1, p4); op.vector(p2, p5); op.vector(p4, p7);
+                    op.vector(p4, p2); op.vector(p6, p4); op.vector(p4, p2);
+                    *(VT*)(dst+j) = p4;
+                }
+
+                limit = width;
+            }
+        }
+    }
+    else if( ksize == 5 )
+    {
+        if( width == 1 || height == 1 )
+        {
+            int len = width + height - 1;
+            int sdelta = height == 1 ? cn : sstep;
+            int sdelta0 = height == 1 ? 0 : sstep - cn;
+            int ddelta = height == 1 ? cn : dstep;
+
+            for( i = 0; i < len; i++, src += sdelta0, dst += ddelta )
+                for( j = 0; j < cn; j++, src++ )
+                {
+                    int i1 = i > 0 ? -sdelta : 0;
+                    int i0 = i > 1 ? -sdelta*2 : i1;
+                    int i3 = i < len-1 ? sdelta : 0;
+                    int i4 = i < len-2 ? sdelta*2 : i3;
+                    T p0 = src[i0], p1 = src[i1], p2 = src[0], p3 = src[i3], p4 = src[i4];
+
+                    op.scalar(p0, p1); op.scalar(p3, p4); op.scalar(p2, p3); op.scalar(p3, p4); op.scalar(p0, p2);
+                    op.scalar(p2, p4); op.scalar(p1, p3); op.scalar(p1, p2);
+                    dst[j] = (T)p2;
+                }
+            return;
+        }
+
+        width *= cn;
+        for( i = 0; i < height; i++, dst += dstep )
+        {
+            const T* row[5];
+            row[0] = src + std::max(i - 2, 0)*sstep;
+            row[1] = src + std::max(i - 1, 0)*sstep;
+            row[2] = src + i*sstep;
+            row[3] = src + std::min(i + 1, height-1)*sstep;
+            row[4] = src + std::min(i + 2, height-1)*sstep;
+            int limit = cn*2;
+
+            for(j = 0;; )
+            {
+                for( ; j < limit; j++ )
+                {
+                    T p[25];
+                    int j1 = j >= cn ? j - cn : j;
+                    int j0 = j >= cn*2 ? j - cn*2 : j1;
+                    int j3 = j < width - cn ? j + cn : j;
+                    int j4 = j < width - cn*2 ? j + cn*2 : j3;
+                    for( k = 0; k < 5; k++ )
+                    {
+                        const T* rowk = row[k];
+                        p[k*5] = rowk[j0]; p[k*5+1] = rowk[j1];
+                        p[k*5+2] = rowk[j]; p[k*5+3] = rowk[j3];
+                        p[k*5+4] = rowk[j4];
+                    }
+
+                    op.scalar(p[1], p[2]); op.scalar(p[0], p[1]); op.scalar(p[1], p[2]); op.scalar(p[4], p[5]); op.scalar(p[3], p[4]);
+                    op.scalar(p[4], p[5]); op.scalar(p[0], p[3]); op.scalar(p[2], p[5]); op.scalar(p[2], p[3]); op.scalar(p[1], p[4]);
+                    op.scalar(p[1], p[2]); op.scalar(p[3], p[4]); op.scalar(p[7], p[8]); op.scalar(p[6], p[7]); op.scalar(p[7], p[8]);
+                    op.scalar(p[10], p[11]); op.scalar(p[9], p[10]); op.scalar(p[10], p[11]); op.scalar(p[6], p[9]); op.scalar(p[8], p[11]);
+                    op.scalar(p[8], p[9]); op.scalar(p[7], p[10]); op.scalar(p[7], p[8]); op.scalar(p[9], p[10]); op.scalar(p[0], p[6]);
+                    op.scalar(p[4], p[10]); op.scalar(p[4], p[6]); op.scalar(p[2], p[8]); op.scalar(p[2], p[4]); op.scalar(p[6], p[8]);
+                    op.scalar(p[1], p[7]); op.scalar(p[5], p[11]); op.scalar(p[5], p[7]); op.scalar(p[3], p[9]); op.scalar(p[3], p[5]);
+                    op.scalar(p[7], p[9]); op.scalar(p[1], p[2]); op.scalar(p[3], p[4]); op.scalar(p[5], p[6]); op.scalar(p[7], p[8]);
+                    op.scalar(p[9], p[10]); op.scalar(p[13], p[14]); op.scalar(p[12], p[13]); op.scalar(p[13], p[14]); op.scalar(p[16], p[17]);
+                    op.scalar(p[15], p[16]); op.scalar(p[16], p[17]); op.scalar(p[12], p[15]); op.scalar(p[14], p[17]); op.scalar(p[14], p[15]);
+                    op.scalar(p[13], p[16]); op.scalar(p[13], p[14]); op.scalar(p[15], p[16]); op.scalar(p[19], p[20]); op.scalar(p[18], p[19]);
+                    op.scalar(p[19], p[20]); op.scalar(p[21], p[22]); op.scalar(p[23], p[24]); op.scalar(p[21], p[23]); op.scalar(p[22], p[24]);
+                    op.scalar(p[22], p[23]); op.scalar(p[18], p[21]); op.scalar(p[20], p[23]); op.scalar(p[20], p[21]); op.scalar(p[19], p[22]);
+                    op.scalar(p[22], p[24]); op.scalar(p[19], p[20]); op.scalar(p[21], p[22]); op.scalar(p[23], p[24]); op.scalar(p[12], p[18]);
+                    op.scalar(p[16], p[22]); op.scalar(p[16], p[18]); op.scalar(p[14], p[20]); op.scalar(p[20], p[24]); op.scalar(p[14], p[16]);
+                    op.scalar(p[18], p[20]); op.scalar(p[22], p[24]); op.scalar(p[13], p[19]); op.scalar(p[17], p[23]); op.scalar(p[17], p[19]);
+                    op.scalar(p[15], p[21]); op.scalar(p[15], p[17]); op.scalar(p[19], p[21]); op.scalar(p[13], p[14]); op.scalar(p[15], p[16]);
+                    op.scalar(p[17], p[18]); op.scalar(p[19], p[20]); op.scalar(p[21], p[22]); op.scalar(p[23], p[24]); op.scalar(p[0], p[12]);
+                    op.scalar(p[8], p[20]); op.scalar(p[8], p[12]); op.scalar(p[4], p[16]); op.scalar(p[16], p[24]); op.scalar(p[12], p[16]);
+                    op.scalar(p[2], p[14]); op.scalar(p[10], p[22]); op.scalar(p[10], p[14]); op.scalar(p[6], p[18]); op.scalar(p[6], p[10]);
+                    op.scalar(p[10], p[12]); op.scalar(p[1], p[13]); op.scalar(p[9], p[21]); op.scalar(p[9], p[13]); op.scalar(p[5], p[17]);
+                    op.scalar(p[13], p[17]); op.scalar(p[3], p[15]); op.scalar(p[11], p[23]); op.scalar(p[11], p[15]); op.scalar(p[7], p[19]);
+                    op.scalar(p[7], p[11]); op.scalar(p[11], p[13]); op.scalar(p[11], p[12]);
+                    dst[j] = (T)p[12];
+                }
+
+                if( limit == width )
+                    break;
+
+                int nlanes = 8 / sizeof(T);
+
+                for( ; (cn % nlanes == 0) && (j <= width - nlanes - cn*2); j += nlanes )
+                {
+                    VT p0 = *(VT*)(row[0]+j-cn*2), p5 = *(VT*)(row[1]+j-cn*2), p10 = *(VT*)(row[2]+j-cn*2), p15 = *(VT*)(row[3]+j-cn*2), p20 = *(VT*)(row[4]+j-cn*2);
+                    VT p1 = *(VT*)(row[0]+j-cn*1), p6 = *(VT*)(row[1]+j-cn*1), p11 = *(VT*)(row[2]+j-cn*1), p16 = *(VT*)(row[3]+j-cn*1), p21 = *(VT*)(row[4]+j-cn*1);
+                    VT p2 = *(VT*)(row[0]+j-cn*0), p7 = *(VT*)(row[1]+j-cn*0), p12 = *(VT*)(row[2]+j-cn*0), p17 = *(VT*)(row[3]+j-cn*0), p22 = *(VT*)(row[4]+j-cn*0);
+                    VT p3 = *(VT*)(row[0]+j+cn*1), p8 = *(VT*)(row[1]+j+cn*1), p13 = *(VT*)(row[2]+j+cn*1), p18 = *(VT*)(row[3]+j+cn*1), p23 = *(VT*)(row[4]+j+cn*1);
+                    VT p4 = *(VT*)(row[0]+j+cn*2), p9 = *(VT*)(row[1]+j+cn*2), p14 = *(VT*)(row[2]+j+cn*2), p19 = *(VT*)(row[3]+j+cn*2), p24 = *(VT*)(row[4]+j+cn*2);
+
+                    op.vector(p1, p2); op.vector(p0, p1); op.vector(p1, p2); op.vector(p4, p5); op.vector(p3, p4);
+                    op.vector(p4, p5); op.vector(p0, p3); op.vector(p2, p5); op.vector(p2, p3); op.vector(p1, p4);
+                    op.vector(p1, p2); op.vector(p3, p4); op.vector(p7, p8); op.vector(p6, p7); op.vector(p7, p8);
+                    op.vector(p10, p11); op.vector(p9, p10); op.vector(p10, p11); op.vector(p6, p9); op.vector(p8, p11);
+                    op.vector(p8, p9); op.vector(p7, p10); op.vector(p7, p8); op.vector(p9, p10); op.vector(p0, p6);
+                    op.vector(p4, p10); op.vector(p4, p6); op.vector(p2, p8); op.vector(p2, p4); op.vector(p6, p8);
+                    op.vector(p1, p7); op.vector(p5, p11); op.vector(p5, p7); op.vector(p3, p9); op.vector(p3, p5);
+                    op.vector(p7, p9); op.vector(p1, p2); op.vector(p3, p4); op.vector(p5, p6); op.vector(p7, p8);
+                    op.vector(p9, p10); op.vector(p13, p14); op.vector(p12, p13); op.vector(p13, p14); op.vector(p16, p17);
+                    op.vector(p15, p16); op.vector(p16, p17); op.vector(p12, p15); op.vector(p14, p17); op.vector(p14, p15);
+                    op.vector(p13, p16); op.vector(p13, p14); op.vector(p15, p16); op.vector(p19, p20); op.vector(p18, p19);
+                    op.vector(p19, p20); op.vector(p21, p22); op.vector(p23, p24); op.vector(p21, p23); op.vector(p22, p24);
+                    op.vector(p22, p23); op.vector(p18, p21); op.vector(p20, p23); op.vector(p20, p21); op.vector(p19, p22);
+                    op.vector(p22, p24); op.vector(p19, p20); op.vector(p21, p22); op.vector(p23, p24); op.vector(p12, p18);
+                    op.vector(p16, p22); op.vector(p16, p18); op.vector(p14, p20); op.vector(p20, p24); op.vector(p14, p16);
+                    op.vector(p18, p20); op.vector(p22, p24); op.vector(p13, p19); op.vector(p17, p23); op.vector(p17, p19);
+                    op.vector(p15, p21); op.vector(p15, p17); op.vector(p19, p21); op.vector(p13, p14); op.vector(p15, p16);
+                    op.vector(p17, p18); op.vector(p19, p20); op.vector(p21, p22); op.vector(p23, p24); op.vector(p0, p12);
+                    op.vector(p8, p20); op.vector(p8, p12); op.vector(p4, p16); op.vector(p16, p24); op.vector(p12, p16);
+                    op.vector(p2, p14); op.vector(p10, p22); op.vector(p10, p14); op.vector(p6, p18); op.vector(p6, p10);
+                    op.vector(p10, p12); op.vector(p1, p13); op.vector(p9, p21); op.vector(p9, p13); op.vector(p5, p17);
+                    op.vector(p13, p17); op.vector(p3, p15); op.vector(p11, p23); op.vector(p11, p15); op.vector(p7, p19);
+                    op.vector(p7, p11); op.vector(p11, p13); op.vector(p11, p12);
+                    *(VT*)(dst+j) = p12;
+                }
+
+                limit = width;
+            }
+        }
+    }
+}
+
+int medianBlur(const uchar* src_data, size_t src_step,
+    uchar* dst_data, size_t dst_step,
+    int width, int height, int depth, int cn, int ksize)
+{
+    bool useSortNet = ((ksize == 3) || (ksize == 5 && ( depth > CV_8U || cn == 2 || cn > 4 )));
+
+    if( useSortNet )
+    {
+        uchar* src_data_rep;
+        if( dst_data == src_data ) {
+            std::vector<uchar> src_data_copy(src_step * height);
+            memcpy(src_data_copy.data(), src_data, src_step * height);
+            src_data_rep = &src_data_copy[0];
+        }
+        else {
+            src_data_rep = (uchar*)src_data;
+        }
+
+        if( depth == CV_8U )
+            medianBlur_SortNet<uchar, int, uint8x8_t>( src_data_rep, src_step, dst_data, dst_step, width, height, cn, ksize );
+        else if( depth == CV_8S )
+            medianBlur_SortNet<schar, int, int8x8_t>( src_data_rep, src_step, dst_data, dst_step, width, height, cn, ksize );
+        else if( depth == CV_16U )
+            medianBlur_SortNet<ushort, int, uint16x4_t>( src_data_rep, src_step, dst_data, dst_step, width, height, cn, ksize );
+        else if( depth == CV_16S )
+            medianBlur_SortNet<short, int, int16x4_t>( src_data_rep, src_step, dst_data, dst_step, width, height, cn, ksize );
+        else 
+            return CV_HAL_ERROR_NOT_IMPLEMENTED;
+
+        return CV_HAL_ERROR_OK;
+    }
+    else return CV_HAL_ERROR_NOT_IMPLEMENTED;
+}
+
+} // namespace ndsrvp
+
+} // namespace cv