OpenCV代码提取 resize函数的实现

 之前在http://blog.csdn.net/fengbingchun/article/details/17335477 中有过对cv::resize函数五种插值算法的介绍。这里将OpenCV3.1中五种插值算法的代码进行了提取调整。支持N通道uchar和float类型。经测试，与OpenCV3.1结果完全一致。

 实现代码resize.hpp：

// fbc_cv is free software and uses the same licence as OpenCV// Email: fengbingchun@163.com#ifndef FBC_CV_RESIZE_HPP_#define FBC_CV_RESIZE_HPP_/* reference: imgproc/include/opencv2/imgproc.hpp              imgproc/src/imgwarp.cpp*/#include "core/mat.hpp"#include "core/base.hpp"#include "core/saturate.hpp"#include "core/utility.hpp"#include "imgproc.hpp"namespace fbc {static const int MAX_ESIZE = 16;// interpolation formulas and tablesconst int INTER_RESIZE_COEF_BITS = 11;const int INTER_RESIZE_COEF_SCALE = 1 << INTER_RESIZE_COEF_BITS;template<typename _Tp, int chs> static int resize_nearest(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst);template<typename _Tp, int chs> static int resize_linear(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst);template<typename _Tp, int chs> static int resize_cubic(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst);template<typename _Tp, int chs> static int resize_area(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst);template<typename _Tp, int chs> static int resize_lanczos4(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst);// resize the image src down to or up to the specified size// support type: uchar/floattemplate<typename _Tp, int chs>int resize(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst, int interpolation = NTER_LINEAR){ FBC_Assert((interpolation >= 0) && (interpolation < 5)); FBC_Assert((src.rows >= 4 && src.cols >= 4) && (dst.rows >= 4  && dst.cols >= 4)); FBC_Assert((sizeof(_Tp) == 1) || sizeof(_Tp) == 4); // uchar || float Size ssize = src.size(); Size dsize = dst.size(); if (dsize == ssize) {  // Source and destination are of same size. Use simple copy.  src.copyTo(dst);  return 0; } switch (interpolation) {  case 0: {   resize_nearest(src, dst);   break;  }  case 1: {   resize_linear(src, dst);   break;  }  case 2: {   resize_cubic(src, dst);   break;  }  case 3: {   resize_area(src, dst);   break;  }  case 4: {   resize_lanczos4(src, dst);   break;  }  default:   return -1; } return 0;}struct DecimateAlpha{ int si, di; float alpha;};template<typename type>static int computeResizeAreaTab(int ssize, int dsize, int cn, double scale, DecimateAlpha* tab){ int k = 0; for (int dx = 0; dx < dsize; dx++) {  double fsx1 = dx * scale;  double fsx2 = fsx1 + scale;  double cellWidth = std::min(scale, ssize - fsx1);  int sx1 = fbcCeil(fsx1), sx2 = fbcFloor(fsx2);  sx2 = std::min(sx2, ssize - 1);  sx1 = std::min(sx1, sx2);  if (sx1 - fsx1 > 1e-3) {   assert(k < ssize * 2);   tab[k].di = dx * cn;   tab[k].si = (sx1 - 1) * cn;   tab[k++].alpha = (float)((sx1 - fsx1) / cellWidth);  }  for (int sx = sx1; sx < sx2; sx++) {   assert(k < ssize * 2);   tab[k].di = dx * cn;   tab[k].si = sx * cn;   tab[k++].alpha = float(1.0 / cellWidth);  }  if (fsx2 - sx2 > 1e-3) {   assert(k < ssize * 2);   tab[k].di = dx * cn;   tab[k].si = sx2 * cn;   tab[k++].alpha = (float)(std::min(std::min(fsx2 - sx2, 1.), cellWidth) / cellWidth);  } } return k;}template<typename ST, typename DT> struct Cast{ typedef ST type1; typedef DT rtype; DT operator()(ST val) const { return saturate_cast<DT>(val); }};template<typename ST, typename DT, int bits> struct FixedPtCast{ typedef ST type1; typedef DT rtype; enum { SHIFT = bits, DELTA = 1 << (bits - 1) }; DT operator()(ST val) const { return saturate_cast<DT>((val + DELTA) >> SHIFT); }};template<typename type>static type clip(type x, type a, type b){ return x >= a ? (x < b ? x : b - 1) : a;}template<typename T, typename WT, typename AT>struct HResizeLinear{ typedef T value_type; typedef WT buf_type; typedef AT alpha_type; void operator()(const T** src, WT** dst, int count,  const int* xofs, const AT* alpha,  int swidth, int dwidth, int cn, int xmin, int xmax, int ONE) const {  int dx, k;  int dx0 = 0;  for (k = 0; k <= count - 2; k++) {   const T *S0 = src[k], *S1 = src[k + 1];   WT *D0 = dst[k], *D1 = dst[k + 1];   for (dx = dx0; dx < xmax; dx++) {    int sx = xofs[dx];    WT a0 = alpha[dx * 2], a1 = alpha[dx * 2 + 1];    WT t0 = S0[sx] * a0 + S0[sx + cn] * a1;    WT t1 = S1[sx] * a0 + S1[sx + cn] * a1;    D0[dx] = t0; D1[dx] = t1;   }   for (; dx < dwidth; dx++) {    int sx = xofs[dx];    D0[dx] = WT(S0[sx] * ONE); D1[dx] = WT(S1[sx] * ONE);   }  }  for (; k < count; k++) {   const T *S = src[k];   WT *D = dst[k];   for (dx = 0; dx < xmax; dx++) {    int sx = xofs[dx];    D[dx] = S[sx] * alpha[dx * 2] + S[sx + cn] * alpha[dx * 2 + 1];   }   for (; dx < dwidth; dx++) {    D[dx] = WT(S[xofs[dx]] * ONE);   }  } }};template<typename T, typename WT, typename AT, class CastOp>struct VResizeLinear{ typedef T value_type; typedef WT buf_type; typedef AT alpha_type; void operator()(const WT** src, T* dst, const AT* beta, int width) const {  WT b0 = beta[0], b1 = beta[1];  const WT *S0 = src[0], *S1 = src[1];  CastOp castOp;  int x = 0;  for (; x <= width - 4; x += 4) {   WT t0, t1;   t0 = S0[x] * b0 + S1[x] * b1;   t1 = S0[x + 1] * b0 + S1[x + 1] * b1;   dst[x] = castOp(t0); dst[x + 1] = castOp(t1);   t0 = S0[x + 2] * b0 + S1[x + 2] * b1;   t1 = S0[x + 3] * b0 + S1[x + 3] * b1;   dst[x + 2] = castOp(t0); dst[x + 3] = castOp(t1);  }  for (; x < width; x++) {   dst[x] = castOp(S0[x] * b0 + S1[x] * b1);  } }};template<>struct VResizeLinear<uchar, int, short, FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS * 2>>{ typedef uchar value_type; typedef int buf_type; typedef short alpha_type; void operator()(const buf_type** src, value_type* dst, const alpha_type* beta, int width) const {  alpha_type b0 = beta[0], b1 = beta[1];  const buf_type *S0 = src[0], *S1 = src[1];  int x = 0;  for (; x <= width - 4; x += 4) {   dst[x + 0] = uchar((((b0 * (S0[x + 0] >> 4)) >> 16) + ((b1 * (S1[x + 0] >> 4)) >> 16) + 2) >> 2);   dst[x + 1] = uchar((((b0 * (S0[x + 1] >> 4)) >> 16) + ((b1 * (S1[x + 1] >> 4)) >> 16) + 2) >> 2);   dst[x + 2] = uchar((((b0 * (S0[x + 2] >> 4)) >> 16) + ((b1 * (S1[x + 2] >> 4)) >> 16) + 2) >> 2);   dst[x + 3] = uchar((((b0 * (S0[x + 3] >> 4)) >> 16) + ((b1 * (S1[x + 3] >> 4)) >> 16) + 2) >> 2);  }  for (; x < width; x++) {   dst[x] = uchar((((b0 * (S0[x] >> 4)) >> 16) + ((b1 * (S1[x] >> 4)) >> 16) + 2) >> 2);  } }};template<typename T, typename WT, typename AT>struct HResizeCubic{ typedef T value_type; typedef WT buf_type; typedef AT alpha_type; void operator()(const T** src, WT** dst, int count,  const int* xofs, const AT* alpha,  int swidth, int dwidth, int cn, int xmin, int xmax) const {  for (int k = 0; k < count; k++) {   const T *S = src[k];   WT *D = dst[k];   int dx = 0, limit = xmin;   for (;;) {    for (; dx < limit; dx++, alpha += 4) {     int j, sx = xofs[dx] - cn;     WT v = 0;     for (j = 0; j < 4; j++) {      int sxj = sx + j*cn;      if ((unsigned)sxj >= (unsigned)swidth) {       while (sxj < 0)        sxj += cn;       while (sxj >= swidth)        sxj -= cn;      }      v += S[sxj] * alpha[j];     }     D[dx] = v;    }    if (limit == dwidth)     break;    for (; dx < xmax; dx++, alpha += 4) {     int sx = xofs[dx];     D[dx] = S[sx - cn] * alpha[0] + S[sx] * alpha[1] +      S[sx + cn] * alpha[2] + S[sx + cn * 2] * alpha[3];    }    limit = dwidth;   }   alpha -= dwidth * 4;  } }};template<typename T, typename WT, typename AT, class CastOp>struct VResizeCubic{ typedef T value_type; typedef WT buf_type; typedef AT alpha_type; void operator()(const WT** src, T* dst, const AT* beta, int width) const {  WT b0 = beta[0], b1 = beta[1], b2 = beta[2], b3 = beta[3];  const WT *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3];  CastOp castOp;  int x = 0;  for (; x < width; x++) {   dst[x] = castOp(S0[x] * b0 + S1[x] * b1 + S2[x] * b2 + S3[x] * b3);  } }};template<typename T, typename WT, typename AT>struct HResizeLanczos4{ typedef T value_type; typedef WT buf_type; typedef AT alpha_type; void operator()(const T** src, WT** dst, int count,  const int* xofs, const AT* alpha,  int swidth, int dwidth, int cn, int xmin, int xmax) const {  for (int k = 0; k < count; k++) {   const T *S = src[k];   WT *D = dst[k];   int dx = 0, limit = xmin;   for (;;) {    for (; dx < limit; dx++, alpha += 8) {     int j, sx = xofs[dx] - cn * 3;     WT v = 0;     for (j = 0; j < 8; j++) {      int sxj = sx + j*cn;      if ((unsigned)sxj >= (unsigned)swidth) {       while (sxj < 0)        sxj += cn;       while (sxj >= swidth)        sxj -= cn;      }      v += S[sxj] * alpha[j];     }     D[dx] = v;    }    if (limit == dwidth)     break;    for (; dx < xmax; dx++, alpha += 8) {     int sx = xofs[dx];     D[dx] = S[sx - cn * 3] * alpha[0] + S[sx - cn * 2] * alpha[1] +      S[sx - cn] * alpha[2] + S[sx] * alpha[3] +      S[sx + cn] * alpha[4] + S[sx + cn * 2] * alpha[5] +      S[sx + cn * 3] * alpha[6] + S[sx + cn * 4] * alpha[7];    }    limit = dwidth;   }   alpha -= dwidth * 8;  } }};template<typename T, typename WT, typename AT, class CastOp>struct VResizeLanczos4{ typedef T value_type; typedef WT buf_type; typedef AT alpha_type; void operator()(const WT** src, T* dst, const AT* beta, int width) const {  CastOp castOp;  int k, x = 0;  for (; x <= width - 4; x += 4) {   WT b = beta[0];   const WT* S = src[0];   WT s0 = S[x] * b, s1 = S[x + 1] * b, s2 = S[x + 2] * b, s3 = S[x + 3] * b;   for (k = 1; k < 8; k++) {    b = beta[k]; S = src[k];    s0 += S[x] * b; s1 += S[x + 1] * b;    s2 += S[x + 2] * b; s3 += S[x + 3] * b;   }   dst[x] = castOp(s0); dst[x + 1] = castOp(s1);   dst[x + 2] = castOp(s2); dst[x + 3] = castOp(s3);  }  for (; x < width; x++) {   dst[x] = castOp(src[0][x] * beta[0] + src[1][x] * beta[1] +    src[2][x] * beta[2] + src[3][x] * beta[3] + src[4][x] * beta[4] +    src[5][x] * beta[5] + src[6][x] * beta[6] + src[7][x] * beta[7]);  } }};template<typename T>struct ResizeAreaFastVec{ ResizeAreaFastVec(int _scale_x, int _scale_y, int _cn, int _step) :  scale_x(_scale_x), scale_y(_scale_y), cn(_cn), step(_step) {  fast_mode = scale_x == 2 && scale_y == 2 && (cn == 1 || cn == 3 || cn == 4); } int operator() (const T* S, T* D, int w) const {  if (!fast_mode) {   return 0;  }  const T* nextS = (const T*)((const uchar*)S + step);  int dx = 0;  if (cn == 1) {   for (; dx < w; ++dx) {    int index = dx * 2;    D[dx] = (T)((S[index] + S[index + 1] + nextS[index] + nextS[index + 1] + 2) >> 2);   }  }  else if (cn == 3) {   for (; dx < w; dx += 3) {    int index = dx * 2;    D[dx] = (T)((S[index] + S[index + 3] + nextS[index] + nextS[index + 3] + 2) >> 2);    D[dx + 1] = (T)((S[index + 1] + S[index + 4] + nextS[index + 1] + nextS[index + 4] + 2) >> 2);    D[dx + 2] = (T)((S[index + 2] + S[index + 5] + nextS[index + 2] + nextS[index + 5] + 2) >> 2);   }  } else {   FBC_Assert(cn == 4);   for (; dx < w; dx += 4) {    int index = dx * 2;    D[dx] = (T)((S[index] + S[index + 4] + nextS[index] + nextS[index + 4] + 2) >> 2);    D[dx + 1] = (T)((S[index + 1] + S[index + 5] + nextS[index + 1] + nextS[index + 5] + 2) >> 2);    D[dx + 2] = (T)((S[index + 2] + S[index + 6] + nextS[index + 2] + nextS[index + 6] + 2) >> 2);    D[dx + 3] = (T)((S[index + 3] + S[index + 7] + nextS[index + 3] + nextS[index + 7] + 2) >> 2);   }  }  return dx; }private: int scale_x, scale_y; int cn; bool fast_mode; int step;};template<typename _Tp, typename value_type, typename buf_type, typename alpha_type, int chs>static void resizeGeneric_Linear(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst, const int* xofs, const void* _alpha, const int* yofs, const void* _beta, int xmin, int xmax, int ksize, int ONE){ Size ssize = src.size(), dsize = dst.size(); int dy, cn = src.channels; ssize.width *= cn; dsize.width *= cn; xmin *= cn; xmax *= cn; // image resize is a separable operation. In case of not too strong Range range(0, dsize.height); int bufstep = (int)alignSize(dsize.width, 16); AutoBuffer<buf_type> _buffer(bufstep*ksize); const value_type* srows[MAX_ESIZE] = { 0 }; buf_type* rows[MAX_ESIZE] = { 0 }; int prev_sy[MAX_ESIZE]; for (int k = 0; k < ksize; k++) {  prev_sy[k] = -1;  rows[k] = (buf_type*)_buffer + bufstep*k; } const alpha_type* beta = (const alpha_type*)_beta + ksize * range.start; HResizeLinear<value_type, buf_type, alpha_type> hresize; VResizeLinear<value_type, buf_type, alpha_type, FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS * 2>> vresize1; VResizeLinear<value_type, buf_type, alpha_type, Cast<float, float>> vresize2; for (dy = range.start; dy < range.end; dy++, beta += ksize) {  int sy0 = yofs[dy], k0 = ksize, k1 = 0, ksize2 = ksize / 2;  for (int k = 0; k < ksize; k++) {   int sy = clip<int>(sy0 - ksize2 + 1 + k, 0, ssize.height);   for (k1 = std::max(k1, k); k1 < ksize; k1++) {    if (sy == prev_sy[k1]) { // if the sy-th row has been computed already, reuse it.     if (k1 > k) {      memcpy(rows[k], rows[k1], bufstep*sizeof(rows[0][0]));     }     break;    }   }   if (k1 == ksize) {    k0 = std::min(k0, k); // remember the first row that needs to be computed   }   srows[k] = (const value_type*)src.ptr(sy);   prev_sy[k] = sy;  }  if (k0 < ksize) {   hresize((const value_type**)(srows + k0), (buf_type**)(rows + k0), ksize - k0, xofs, (const alpha_type*)(_alpha),    ssize.width, dsize.width, cn, xmin, xmax, ONE);  }  if (sizeof(_Tp) == 1) { // uchar   vresize1((const buf_type**)rows, (value_type*)(dst.data + dst.step*dy), beta, dsize.width);  } else { // float   vresize2((const buf_type**)rows, (value_type*)(dst.data + dst.step*dy), beta, dsize.width);  } }}template<typename _Tp, typename value_type, typename buf_type, typename alpha_type, int chs>static void resizeGeneric_Cubic(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst, const int* xofs, const void* _alpha, const int* yofs, const void* _beta, int xmin, int xmax, int ksize){ Size ssize = src.size(), dsize = dst.size(); int dy, cn = src.channels; ssize.width *= cn; dsize.width *= cn; xmin *= cn; xmax *= cn; // image resize is a separable operation. In case of not too strong Range range(0, dsize.height); int bufstep = (int)alignSize(dsize.width, 16); AutoBuffer<buf_type> _buffer(bufstep*ksize); const value_type* srows[MAX_ESIZE] = { 0 }; buf_type* rows[MAX_ESIZE] = { 0 }; int prev_sy[MAX_ESIZE]; for (int k = 0; k < ksize; k++) {  prev_sy[k] = -1;  rows[k] = (buf_type*)_buffer + bufstep*k; } const alpha_type* beta = (const alpha_type*)_beta + ksize * range.start; HResizeCubic<value_type, buf_type, alpha_type> hresize; VResizeCubic<value_type, buf_type, alpha_type, FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS * 2>> vresize1; VResizeCubic<value_type, buf_type, alpha_type, Cast<float, float>> vresize2; for (dy = range.start; dy < range.end; dy++, beta += ksize) {  int sy0 = yofs[dy], k0 = ksize, k1 = 0, ksize2 = ksize / 2;  for (int k = 0; k < ksize; k++) {   int sy = clip<int>(sy0 - ksize2 + 1 + k, 0, ssize.height);   for (k1 = std::max(k1, k); k1 < ksize; k1++) {    if (sy == prev_sy[k1]) { // if the sy-th row has been computed already, reuse it.     if (k1 > k) {      memcpy(rows[k], rows[k1], bufstep*sizeof(rows[0][0]));     }     break;    }   }   if (k1 == ksize) {    k0 = std::min(k0, k); // remember the first row that needs to be computed   }   srows[k] = (const value_type*)src.ptr(sy);   prev_sy[k] = sy;  }  if (k0 < ksize) {   hresize((const value_type**)(srows + k0), (buf_type**)(rows + k0), ksize - k0, xofs, (const alpha_type*)(_alpha),    ssize.width, dsize.width, cn, xmin, xmax);  }  if (sizeof(_Tp) == 1) { // uchar   vresize1((const buf_type**)rows, (value_type*)(dst.data + dst.step*dy), beta, dsize.width);  } else { // float   vresize2((const buf_type**)rows, (value_type*)(dst.data + dst.step*dy), beta, dsize.width);  } }}template<typename _Tp, typename value_type, typename buf_type, typename alpha_type, int chs>static void resizeGeneric_Lanczos4(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst, const int* xofs, const void* _alpha, const int* yofs, const void* _beta, int xmin, int xmax, int ksize){ Size ssize = src.size(), dsize = dst.size(); int dy, cn = src.channels; ssize.width *= cn; dsize.width *= cn; xmin *= cn; xmax *= cn; // image resize is a separable operation. In case of not too strong Range range(0, dsize.height); int bufstep = (int)alignSize(dsize.width, 16); AutoBuffer<buf_type> _buffer(bufstep*ksize); const value_type* srows[MAX_ESIZE] = { 0 }; buf_type* rows[MAX_ESIZE] = { 0 }; int prev_sy[MAX_ESIZE]; for (int k = 0; k < ksize; k++) {  prev_sy[k] = -1;  rows[k] = (buf_type*)_buffer + bufstep*k; } const alpha_type* beta = (const alpha_type*)_beta + ksize * range.start; HResizeLanczos4<value_type, buf_type, alpha_type> hresize; VResizeLanczos4<value_type, buf_type, alpha_type, FixedPtCast<int, uchar, INTER_RESIZE_COEF_BITS * 2>> vresize1; VResizeLanczos4<value_type, buf_type, alpha_type, Cast<float, float>> vresize2; for (dy = range.start; dy < range.end; dy++, beta += ksize) {  int sy0 = yofs[dy], k0 = ksize, k1 = 0, ksize2 = ksize / 2;  for (int k = 0; k < ksize; k++) {   int sy = clip<int>(sy0 - ksize2 + 1 + k, 0, ssize.height);   for (k1 = std::max(k1, k); k1 < ksize; k1++) {    if (sy == prev_sy[k1]) { // if the sy-th row has been computed already, reuse it.     if (k1 > k) {      memcpy(rows[k], rows[k1], bufstep*sizeof(rows[0][0]));     }     break;    }   }   if (k1 == ksize) {    k0 = std::min(k0, k); // remember the first row that needs to be computed   }   srows[k] = (const value_type*)src.ptr(sy);   prev_sy[k] = sy;  }  if (k0 < ksize) {   hresize((const value_type**)(srows + k0), (buf_type**)(rows + k0), ksize - k0, xofs, (const alpha_type*)(_alpha),    ssize.width, dsize.width, cn, xmin, xmax);  }  if (sizeof(_Tp) == 1) { // uchar   vresize1((const buf_type**)rows, (value_type*)(dst.data + dst.step*dy), beta, dsize.width);  }  else { // float   vresize2((const buf_type**)rows, (value_type*)(dst.data + dst.step*dy), beta, dsize.width);  } }}template<typename _Tp, typename T, typename WT, int chs>static void resizeGeneric_Area(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst, const DecimateAlpha* xtab0, int xtab_size0, const DecimateAlpha* ytab, int ytab_size, const int* tabofs){ Size dsize = dst.size(); int cn = dst.channels; Range range(0, dsize.height); dsize.width *= cn; AutoBuffer<WT> _buffer(dsize.width * 2); const DecimateAlpha* xtab = xtab0; int xtab_size = xtab_size0; WT *buf = _buffer, *sum = buf + dsize.width; int j_start = tabofs[range.start], j_end = tabofs[range.end], j, k, dx, prev_dy = ytab[j_start].di; for (dx = 0; dx < dsize.width; dx++) {  sum[dx] = (WT)0; } for (j = j_start; j < j_end; j++) {  WT beta = ytab[j].alpha;  int dy = ytab[j].di;  int sy = ytab[j].si;  const T* S = (const T*)src.ptr(sy);  for (dx = 0; dx < dsize.width; dx++) {   buf[dx] = (WT)0;  }  if (cn == 1) {   for (k = 0; k < xtab_size; k++) {    int dxn = xtab[k].di;    WT alpha = xtab[k].alpha;    buf[dxn] += S[xtab[k].si] * alpha;   }  } else if (cn == 2) {   for (k = 0; k < xtab_size; k++) {    int sxn = xtab[k].si;    int dxn = xtab[k].di;    WT alpha = xtab[k].alpha;    WT t0 = buf[dxn] + S[sxn] * alpha;    WT t1 = buf[dxn + 1] + S[sxn + 1] * alpha;    buf[dxn] = t0; buf[dxn + 1] = t1;   }  } else if (cn == 3) {   for (k = 0; k < xtab_size; k++) {    int sxn = xtab[k].si;    int dxn = xtab[k].di;    WT alpha = xtab[k].alpha;    WT t0 = buf[dxn] + S[sxn] * alpha;    WT t1 = buf[dxn + 1] + S[sxn + 1] * alpha;    WT t2 = buf[dxn + 2] + S[sxn + 2] * alpha;    buf[dxn] = t0; buf[dxn + 1] = t1; buf[dxn + 2] = t2;   }  } else if (cn == 4) {   for (k = 0; k < xtab_size; k++) {    int sxn = xtab[k].si;    int dxn = xtab[k].di;    WT alpha = xtab[k].alpha;    WT t0 = buf[dxn] + S[sxn] * alpha;    WT t1 = buf[dxn + 1] + S[sxn + 1] * alpha;    buf[dxn] = t0; buf[dxn + 1] = t1;    t0 = buf[dxn + 2] + S[sxn + 2] * alpha;    t1 = buf[dxn + 3] + S[sxn + 3] * alpha;    buf[dxn + 2] = t0; buf[dxn + 3] = t1;   }  } else {   for (k = 0; k < xtab_size; k++) {    int sxn = xtab[k].si;    int dxn = xtab[k].di;    WT alpha = xtab[k].alpha;    for (int c = 0; c < cn; c++)     buf[dxn + c] += S[sxn + c] * alpha;   }  }  if (dy != prev_dy) {   T* D = (T*)dst.ptr(prev_dy);   for (dx = 0; dx < dsize.width; dx++) {    D[dx] = saturate_cast<T>(sum[dx]);    sum[dx] = beta*buf[dx];   }   prev_dy = dy;  } else {   for (dx = 0; dx < dsize.width; dx++) {    sum[dx] += beta*buf[dx];   }  } } T* D = (T*)dst.ptr(prev_dy); for (dx = 0; dx < dsize.width; dx++) {  D[dx] = saturate_cast<T>(sum[dx]); }}template<typename _Tp, typename T, typename WT, int chs>static void resizeGeneric_AreaFast(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst, const int* ofs, const int* xofs, int scale_x, int scale_y){ Size ssize = src.size(), dsize = dst.size(); int cn = src.channels; Range range(0, dsize.height); int area = scale_x*scale_y; float scale = 1.f / (area); int dwidth1 = (ssize.width / scale_x)*cn; dsize.width *= cn; ssize.width *= cn; int dy, dx, k = 0; ResizeAreaFastVec<uchar> vop(scale_x, scale_y, src.channels, (int)src.step); for (dy = range.start; dy < range.end; dy++) {  T* D = (T*)(dst.data + dst.step*dy);  int sy0 = dy*scale_y;  int w = sy0 + scale_y <= ssize.height ? dwidth1 : 0;  if (sy0 >= ssize.height) {   for (dx = 0; dx < dsize.width; dx++) {    D[dx] = 0;   }   continue;  }  dx = sizeof(_Tp) == 1 ? vop(src.ptr(sy0), (uchar*)D, w) : 0;  for (; dx < w; dx++) {   const T* S = (const T*)src.ptr(sy0) +xofs[dx];   WT sum = 0;   k = 0;   for (; k <= area - 4; k += 4) {    sum += S[ofs[k]] + S[ofs[k + 1]] + S[ofs[k + 2]] + S[ofs[k + 3]];   }   for (; k < area; k++) {    sum += S[ofs[k]];   }   D[dx] = saturate_cast<T>(sum * scale);  }  for (; dx < dsize.width; dx++) {   WT sum = 0;   int count = 0, sx0 = xofs[dx];   if (sx0 >= ssize.width) {    D[dx] = 0;   }   for (int sy = 0; sy < scale_y; sy++) {    if (sy0 + sy >= ssize.height) {     break;    }    const T* S = (const T*)src.ptr(sy0 + sy) + sx0;    for (int sx = 0; sx < scale_x*cn; sx += cn) {     if (sx0 + sx >= ssize.width) {      break;     }     sum += S[sx];     count++;    }   }   D[dx] = saturate_cast<T>((float)sum / count);  } }}template<typename _Tp>static void interpolateCubic(_Tp x, _Tp* coeffs){ const float A = -0.75f; coeffs[0] = ((A*(x + 1) - 5 * A)*(x + 1) + 8 * A)*(x + 1) - 4 * A; coeffs[1] = ((A + 2)*x - (A + 3))*x*x + 1; coeffs[2] = ((A + 2)*(1 - x) - (A + 3))*(1 - x)*(1 - x) + 1; coeffs[3] = 1.f - coeffs[0] - coeffs[1] - coeffs[2];}template<typename _Tp>static void interpolateLanczos4(_Tp x, _Tp* coeffs){ static const double s45 = 0.70710678118654752440084436210485; static const double cs[][2] = { { 1, 0 }, { -s45, -s45 }, { 0, 1 }, { s45, -s45 }, { -1, 0 }, { s45, s45 }, { 0, -1 }, { -s45, s45 } }; if (x < FLT_EPSILON) {  for (int i = 0; i < 8; i++) {   coeffs[i] = 0;  }  coeffs[3] = 1;  return; } float sum = 0; double y0 = -(x + 3)*FBC_PI*0.25, s0 = sin(y0), c0 = cos(y0); for (int i = 0; i < 8; i++) {  double y = -(x + 3 - i)*FBC_PI*0.25;  coeffs[i] = (float)((cs[i][0] * s0 + cs[i][1] * c0) / (y*y));  sum += coeffs[i]; } sum = 1.f / sum; for (int i = 0; i < 8; i++) {  coeffs[i] *= sum; }}template<typename _Tp, int chs>static int resize_nearest(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst){ Size ssize = src.size(); Size dsize = dst.size(); double fx = (double)dsize.width / ssize.width; double fy = (double)dsize.height / ssize.height; AutoBuffer<int> _x_ofs(dsize.width); int* x_ofs = _x_ofs; int pix_size = (int)src.elemSize(); int pix_size4 = (int)(pix_size / sizeof(int)); double ifx = 1. / fx, ify = 1. / fy; for (int x = 0; x < dsize.width; x++) {  int sx = fbcFloor(x*ifx);  x_ofs[x] = std::min(sx, ssize.width - 1)*pix_size; } Range range(0, dsize.height); int x, y; for (y = range.start; y < range.end; y++) {  uchar* D = dst.data + dst.step*y;  int sy = std::min(fbcFloor(y*ify), ssize.height - 1);  const uchar* S = src.ptr(sy);  switch (pix_size) {  case 1:   for (x = 0; x <= dsize.width - 2; x += 2) {    uchar t0 = S[x_ofs[x]];    uchar t1 = S[x_ofs[x + 1]];    D[x] = t0;    D[x + 1] = t1;   }   for (; x < dsize.width; x++) {    D[x] = S[x_ofs[x]];   }   break;  case 2:   for (x = 0; x < dsize.width; x++) {    *(ushort*)(D + x * 2) = *(ushort*)(S + x_ofs[x]);   }   break;  case 3:   for (x = 0; x < dsize.width; x++, D += 3) {    const uchar* _tS = S + x_ofs[x];    D[0] = _tS[0]; D[1] = _tS[1]; D[2] = _tS[2];   }   break;  case 4:   for (x = 0; x < dsize.width; x++) {    *(int*)(D + x * 4) = *(int*)(S + x_ofs[x]);   }   break;  case 6:   for (x = 0; x < dsize.width; x++, D += 6) {    const ushort* _tS = (const ushort*)(S + x_ofs[x]);    ushort* _tD = (ushort*)D;    _tD[0] = _tS[0]; _tD[1] = _tS[1]; _tD[2] = _tS[2];   }   break;  case 8:   for (x = 0; x < dsize.width; x++, D += 8) {    const int* _tS = (const int*)(S + x_ofs[x]);    int* _tD = (int*)D;    _tD[0] = _tS[0]; _tD[1] = _tS[1];   }   break;  case 12:   for (x = 0; x < dsize.width; x++, D += 12) {    const int* _tS = (const int*)(S + x_ofs[x]);    int* _tD = (int*)D;    _tD[0] = _tS[0]; _tD[1] = _tS[1]; _tD[2] = _tS[2];   }   break;  default:   for (x = 0; x < dsize.width; x++, D += pix_size) {    const int* _tS = (const int*)(S + x_ofs[x]);    int* _tD = (int*)D;    for (int k = 0; k < pix_size4; k++)     _tD[k] = _tS[k];   }  } } return 0;}template<typename _Tp, int chs>static int resize_linear(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst){ Size ssize = src.size(); Size dsize = dst.size(); double inv_scale_x = (double)dsize.width / ssize.width; double inv_scale_y = (double)dsize.height / ssize.height; double scale_x = 1. / inv_scale_x, scale_y = 1. / inv_scale_y; int iscale_x = saturate_cast<int>(scale_x); int iscale_y = saturate_cast<int>(scale_y); bool is_area_fast = std::abs(scale_x - iscale_x) < DBL_EPSILON && std::abs(scale_y - iscale_y) < DBL_EPSILON; // in case of scale_x && scale_y is equal to 2 // INTER_AREA (fast) also is equal to INTER_LINEAR if (is_area_fast && iscale_x == 2 && iscale_y == 2) {  resize_area(src, dst);  return 0; } int cn = dst.channels; int k, sx, sy, dx, dy; int xmin = 0, xmax = dsize.width, width = dsize.width*cn; bool fixpt = sizeof(_Tp) == 1 ? true : false; float fx, fy; int ksize = 2, ksize2; ksize2 = ksize / 2; AutoBuffer<uchar> _buffer((width + dsize.height)*(sizeof(int) + sizeof(float)*ksize)); int* xofs = (int*)(uchar*)_buffer; int* yofs = xofs + width; float* alpha = (float*)(yofs + dsize.height); short* ialpha = (short*)alpha; float* beta = alpha + width*ksize; short* ibeta = ialpha + width*ksize; float cbuf[MAX_ESIZE]; for (dx = 0; dx < dsize.width; dx++) {  fx = (float)((dx + 0.5)*scale_x - 0.5);  sx = fbcFloor(fx);  fx -= sx;  if (sx < ksize2 - 1) {   xmin = dx + 1;   if (sx < 0) {    fx = 0, sx = 0;   }  }  if (sx + ksize2 >= ssize.width) {   xmax = std::min(xmax, dx);   if (sx >= ssize.width - 1) {    fx = 0, sx = ssize.width - 1;   }  }  for (k = 0, sx *= cn; k < cn; k++) {   xofs[dx*cn + k] = sx + k;  }  cbuf[0] = 1.f - fx;  cbuf[1] = fx;  if (fixpt) {   for (k = 0; k < ksize; k++) {    ialpha[dx*cn*ksize + k] = saturate_cast<short>(cbuf[k] * INTER_RESIZE_COEF_SCALE);   }   for (; k < cn*ksize; k++) {    ialpha[dx*cn*ksize + k] = ialpha[dx*cn*ksize + k - ksize];   }  } else {   for (k = 0; k < ksize; k++) {    alpha[dx*cn*ksize + k] = cbuf[k];   }   for (; k < cn*ksize; k++) {    alpha[dx*cn*ksize + k] = alpha[dx*cn*ksize + k - ksize];   }  } } for (dy = 0; dy < dsize.height; dy++) {  fy = (float)((dy + 0.5)*scale_y - 0.5);  sy = fbcFloor(fy);  fy -= sy;  yofs[dy] = sy;  cbuf[0] = 1.f - fy;  cbuf[1] = fy;  if (fixpt) {   for (k = 0; k < ksize; k++) {    ibeta[dy*ksize + k] = saturate_cast<short>(cbuf[k] * INTER_RESIZE_COEF_SCALE);   }  } else {   for (k = 0; k < ksize; k++) {    beta[dy*ksize + k] = cbuf[k];   }  } } if (sizeof(_Tp) == 1) { // uchar  typedef uchar value_type; // HResizeLinear/VResizeLinear  typedef int buf_type;  typedef short alpha_type;  int ONE = INTER_RESIZE_COEF_SCALE;  resizeGeneric_Linear<_Tp, value_type, buf_type, alpha_type, chs>(src, dst,   xofs, fixpt ? (void*)ialpha : (void*)alpha, yofs, fixpt ? (void*)ibeta : (void*)beta, xmin, xmax, ksize, ONE); } else if (sizeof(_Tp) == 4) { // float  typedef float value_type; // HResizeLinear/VResizeLinear  typedef float buf_type;  typedef float alpha_type;  int ONE = 1;  resizeGeneric_Linear<_Tp, value_type, buf_type, alpha_type, chs>(src, dst,   xofs, fixpt ? (void*)ialpha : (void*)alpha, yofs, fixpt ? (void*)ibeta : (void*)beta, xmin, xmax, ksize, ONE); } else {  fprintf(stderr, "not support type\n");  return -1; } return 0;}template<typename _Tp, int chs>static int resize_cubic(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst){ Size ssize = src.size(); Size dsize = dst.size(); double inv_scale_x = (double)dsize.width / ssize.width; double inv_scale_y = (double)dsize.height / ssize.height; double scale_x = 1. / inv_scale_x, scale_y = 1. / inv_scale_y; int cn = dst.channels; int k, sx, sy, dx, dy; int xmin = 0, xmax = dsize.width, width = dsize.width*cn; bool fixpt = sizeof(_Tp) == 1 ? true : false; float fx, fy; int ksize = 4, ksize2; ksize2 = ksize / 2; AutoBuffer<uchar> _buffer((width + dsize.height)*(sizeof(int) + sizeof(float)*ksize)); int* xofs = (int*)(uchar*)_buffer; int* yofs = xofs + width; float* alpha = (float*)(yofs + dsize.height); short* ialpha = (short*)alpha; float* beta = alpha + width*ksize; short* ibeta = ialpha + width*ksize; float cbuf[MAX_ESIZE]; for (dx = 0; dx < dsize.width; dx++) {  fx = (float)((dx + 0.5)*scale_x - 0.5);  sx = fbcFloor(fx);  fx -= sx;  if (sx < ksize2 - 1) {   xmin = dx + 1;  }  if (sx + ksize2 >= ssize.width) {   xmax = std::min(xmax, dx);  }  for (k = 0, sx *= cn; k < cn; k++) {   xofs[dx*cn + k] = sx + k;  }  interpolateCubic<float>(fx, cbuf);  if (fixpt) {   for (k = 0; k < ksize; k++) {    ialpha[dx*cn*ksize + k] = saturate_cast<short>(cbuf[k] * INTER_RESIZE_COEF_SCALE);   }   for (; k < cn*ksize; k++) {    ialpha[dx*cn*ksize + k] = ialpha[dx*cn*ksize + k - ksize];   }  } else {   for (k = 0; k < ksize; k++) {    alpha[dx*cn*ksize + k] = cbuf[k];   }   for (; k < cn*ksize; k++) {    alpha[dx*cn*ksize + k] = alpha[dx*cn*ksize + k - ksize];   }  } } for (dy = 0; dy < dsize.height; dy++) {  fy = (float)((dy + 0.5)*scale_y - 0.5);  sy = cvFloor(fy);  fy -= sy;  yofs[dy] = sy;  interpolateCubic<float>(fy, cbuf);  if (fixpt) {   for (k = 0; k < ksize; k++) {    ibeta[dy*ksize + k] = saturate_cast<short>(cbuf[k] * INTER_RESIZE_COEF_SCALE);   }  } else {   for (k = 0; k < ksize; k++) {    beta[dy*ksize + k] = cbuf[k];   }  } } if (sizeof(_Tp) == 1) { // uchar  typedef uchar value_type; // HResizeCubic/VResizeCubic  typedef int buf_type;  typedef short alpha_type;  resizeGeneric_Cubic<_Tp, value_type, buf_type, alpha_type, chs>(src, dst,   xofs, fixpt ? (void*)ialpha : (void*)alpha, yofs, fixpt ? (void*)ibeta : (void*)beta, xmin, xmax, ksize); } else if (sizeof(_Tp) == 4) { // float  typedef float value_type; // HResizeCubic/VResizeCubic  typedef float buf_type;  typedef float alpha_type;  resizeGeneric_Cubic<_Tp, value_type, buf_type, alpha_type, chs>(src, dst,   xofs, fixpt ? (void*)ialpha : (void*)alpha, yofs, fixpt ? (void*)ibeta : (void*)beta, xmin, xmax, ksize); } else {  fprintf(stderr, "not support type\n");  return -1; } return 0;}template<typename _Tp, int chs>static int resize_area(const Mat_<_Tp, chs>& src, Mat_<_Tp, chs>& dst){ Size ssize = src.size(); Size dsize = dst.size(); int cn = dst.channels; double inv_scale_x = (double)dsize.width / ssize.width; double inv_scale_y = (double)dsize.height / ssize.height; double scale_x = 1. / inv_scale_x, scale_y = 1. / inv_scale_y; int iscale_x = saturate_cast<int>(scale_x); int iscale_y = saturate_cast<int>(scale_y); bool is_area_fast = std::abs(scale_x - iscale_x) < DBL_EPSILON && std::abs(scale_y - iscale_y) < DBL_EPSILON; int k, sx, sy, dx, dy; // true "area" interpolation is only implemented for the case (scale_x <= 1 && scale_y <= 1). // In other cases it is emulated using some variant of bilinear interpolation if (scale_x >= 1 && scale_y >= 1) {  if (is_area_fast) {   int area = iscale_x*iscale_y;   size_t srcstep = src.step / sizeof(_Tp);   AutoBuffer<int> _ofs(area + dsize.width*cn);   int* ofs = _ofs;   int* xofs = ofs + area;   for (sy = 0, k = 0; sy < iscale_y; sy++) {    for (sx = 0; sx < iscale_x; sx++) {     ofs[k++] = (int)(sy*srcstep + sx*cn);    }   }   for (dx = 0; dx < dsize.width; dx++) {    int j = dx * cn;    sx = iscale_x * j;    for (k = 0; k < cn; k++) {     xofs[j + k] = sx + k;    }   }   if (sizeof(_Tp) == 1) { // uchar    typedef uchar T;    typedef int WT;    resizeGeneric_AreaFast<_Tp, T, WT, chs>(src, dst, ofs, xofs, iscale_x, iscale_y);   } else if (sizeof(_Tp) == 4) { // float    typedef float T;    typedef float WT;    resizeGeneric_AreaFast<_Tp, T, WT, chs>(src, dst, ofs, xofs, iscale_x, iscale_y);   } else {    fprintf(stderr, "not support type\n");    return -1;   }   return 0;  }  FBC_Assert(cn <= 4);  AutoBuffer<DecimateAlpha> _xytab((ssize.width + ssize.height) * 2);  DecimateAlpha* xtab = _xytab, *ytab = xtab + ssize.width * 2;  int xtab_size = computeResizeAreaTab<int>(ssize.width, dsize.width, cn, scale_x, xtab);  int ytab_size = computeResizeAreaTab<int>(ssize.height, dsize.height, 1, scale_y, ytab);  AutoBuffer<int> _tabofs(dsize.height + 1);  int* tabofs = _tabofs;  for (k = 0, dy = 0; k < ytab_size; k++) {   if (k == 0 || ytab[k].di != ytab[k - 1].di) {    assert(ytab[k].di == dy);    tabofs[dy++] = k;   }  }  tabofs[dy] = ytab_size;  if (sizeof(_Tp) == 1) { // uchar   typedef uchar T;   typedef float WT;   resizeGeneric_Area<_Tp, T, WT, chs>(src, dst, xtab, xtab_size, ytab, ytab_size, tabofs);  } else if (sizeof(_Tp) == 4) { // float   typedef float T;   typedef float WT;   resizeGeneric_Area<_Tp, T, WT, chs>(src, dst, xtab, xtab_size, ytab, ytab_size, tabofs);  } else {   fprintf(stderr, "not support type\n");   return -1;  }  return 0; } int xmin = 0, xmax = dsize.width, width = dsize.width*cn; bool fixpt = sizeof(_Tp) == 1 ? true : false; float fx, fy; int ksize = 2, ksize2; ksize2 = ksize / 2; AutoBuffer<uchar> _buffer((width + dsize.height)*(sizeof(int) + sizeof(float)*ksize)); int* xofs = (int*)(uchar*)_buffer; int* yofs = xofs + width; float* alpha = (float*)(yofs + dsize.height); short* ialpha = (short*)alpha; float* beta = alpha + width*ksize; short* ibeta = ialpha + width*ksize; float cbuf[MAX_ESIZE]; for (dx = 0; dx < dsize.width; dx++) {  sx = fbcFloor(dx*scale_x);  fx = (float)((dx + 1) - (sx + 1)*inv_scale_x);  fx = fx <= 0 ? 0.f : fx - fbcFloor(fx);  if (sx < ksize2 - 1) {   xmin = dx + 1;   if (sx < 0) {    fx = 0, sx = 0;   }  }  if (sx + ksize2 >= ssize.width) {   xmax = std::min(xmax, dx);   if (sx >= ssize.width - 1) {    fx = 0, sx = ssize.width - 1;   }  }  for (k = 0, sx *= cn; k < cn; k++) {   xofs[dx*cn + k] = sx + k;  }  cbuf[0] = 1.f - fx;  cbuf[1] = fx;  if (fixpt) {   for (k = 0; k < ksize; k++) {    ialpha[dx*cn*ksize + k] = saturate_cast<short>(cbuf[k] * INTER_RESIZE_COEF_SCALE);   }   for (; k < cn*ksize; k++) {    ialpha[dx*cn*ksize + k] = ialpha[dx*cn*ksize + k - ksize];   }  } else {   for (k = 0; k < ksize; k++) {    alpha[dx*cn*ksize + k] = cbuf[k];   }   for (; k < cn*ksize; k++) {    alpha[dx*cn*ksize + k] = alpha[dx*cn*ksize + k - ksize];   }  } } for (dy = 0; dy < dsize.height; dy++) {  sy = fbcFloor(dy*scale_y);  fy = (float)((dy + 1) - (sy + 1)*inv_scale_y);  fy = fy <= 0 ? 0.f : fy - fbcFloor(fy);  yofs[dy] = sy;  cbuf[0] = 1.f - fy;  cbuf[1] = fy;  if (fixpt) {   for (k = 0; k < ksize; k++) {    ibeta[dy*ksize + k] = saturate_cast<short>(cbuf[k] * INTER_RESIZE_COEF_SCALE);   }  } else {   for (k = 0; k < ksize; k++) {    beta[dy*ksize + k] = cbuf[k];   }  } } if (sizeof(_Tp) == 1) { // uchar  typedef uchar value_type; // HResizeLinear/VResizeLinear  typedef int buf_type;  typedef short alpha_type;  int ONE = INTER_RESIZE_COEF_SCALE;  resizeGeneric_Linear<_Tp, value_type, buf_type, alpha_type, chs>(src, dst,   xofs, fixpt ? (void*)ialpha : (void*)alpha, yofs, fixpt ? (void*)ibeta : (void*)beta, xmin, xmax, ksize, ONE); } else if (sizeof(_Tp) == 4) { // float  typedef再分享一下我老师大神的人工智能教程吧。零基础！通俗易懂！风趣幽默！还带黄段子！希望你也加入到我们人工智能的队伍中来！https://blog.csdn.net/jiangjunshow