OpenCV中resize函数五种插值算法的实现过程

最新版OpenCV2.4.7中，cv::resize函数有五种插值算法：最近邻、双线性、双三次、基于像素区域关系、兰索斯插值。下面用for循环代替cv::resize函数来说明其详细的插值实现过程，其中部分代码摘自于cv::resize函数中的源代码。

每种插值算法的前部分代码是相同的，如下：

  1. 	cv::Mat matSrc, matDst1, matDst2;
  2.  
  3. 	matSrc = cv::imread("lena.jpg", 2 | 4);
  4. 	matDst1 = cv::Mat(cv::Size(800, 1000), matSrc.type(), cv::Scalar::all(0));
  5. 	matDst2 = cv::Mat(matDst1.size(), matSrc.type(), cv::Scalar::all(0));
  6.  
  7. 	double scale_x = (double)matSrc.cols / matDst1.cols;
  8. 	double scale_y = (double)matSrc.rows / matDst1.rows;

1、最近邻：公式，

                  

 

  1. 	for (int i = 0; i < matDst1.cols; ++i)
  2. 	{
  3. 		int sx = cvFloor(i * scale_x);
  4. 		sx = std::min(sx, matSrc.cols - 1);
  5. 		for (int j = 0; j < matDst1.rows; ++j)
  6. 		{
  7. 			int sy = cvFloor(j * scale_y);
  8. 			sy = std::min(sy, matSrc.rows - 1);
  9. 			matDst1.at(j, i) = matSrc.at(sy, sx);
  10. 		}
  11. 	}
  12. 	cv::imwrite("nearest_1.jpg", matDst1);
  13.  
  14. 	cv::resize(matSrc, matDst2, matDst1.size(), 0, 0, 0);
  15. 	cv::imwrite("nearest_2.jpg", matDst2);

2、双线性：由相邻的四像素(2*2)计算得出，公式，

 

  1. 	uchar* dataDst = matDst1.data;
  2. 	int stepDst = matDst1.step;
  3. 	uchar* dataSrc = matSrc.data;
  4. 	int stepSrc = matSrc.step;
  5. 	int iWidthSrc = matSrc.cols;
  6. 	int iHiehgtSrc = matSrc.rows;
  7.  
  8. 	for (int j = 0; j < matDst1.rows; ++j)
  9. 	{
  10. 		float fy = (float)((j + 0.5) * scale_y - 0.5);
  11. 		int sy = cvFloor(fy);
  12. 		fy -= sy;
  13. 		sy = std::min(sy, iHiehgtSrc - 2);
  14. 		sy = std::max(0, sy);
  15.  
  16. 		short cbufy[2];
  17. 		cbufy[0] = cv::saturate_cast<short>((1.f - fy) * 2048);
  18. 		cbufy[1] = 2048 - cbufy[0];
  19.  
  20. 		for (int i = 0; i < matDst1.cols; ++i)
  21. 		{
  22. 			float fx = (float)((i + 0.5) * scale_x - 0.5);
  23. 			int sx = cvFloor(fx);
  24. 			fx -= sx;
  25.  
  26. 			if (sx < 0) {
  27. 				fx = 0, sx = 0;
  28. 			}
  29. 			if (sx >= iWidthSrc - 1) {
  30. 				fx = 0, sx = iWidthSrc - 2;
  31. 			}
  32.  
  33. 			short cbufx[2];
  34. 			cbufx[0] = cv::saturate_cast<short>((1.f - fx) * 2048);
  35. 			cbufx[1] = 2048 - cbufx[0];
  36.  
  37. 			for (int k = 0; k < matSrc.channels(); ++k)
  38. 			{
  39. 				*(dataDst+ j*stepDst + 3*i + k) = (*(dataSrc + sy*stepSrc + 3*sx + k) * cbufx[0] * cbufy[0] + 
  40. 					*(dataSrc + (sy+1)*stepSrc + 3*sx + k) * cbufx[0] * cbufy[1] + 
  41. 					*(dataSrc + sy*stepSrc + 3*(sx+1) + k) * cbufx[1] * cbufy[0] + 
  42. 					*(dataSrc + (sy+1)*stepSrc + 3*(sx+1) + k) * cbufx[1] * cbufy[1]) >> 22;
  43. 			}
  44. 		}
  45. 	}
  46. 	cv::imwrite("linear_1.jpg", matDst1);
  47.  
  48. 	cv::resize(matSrc, matDst2, matDst1.size(), 0, 0, 1);
  49. 	cv::imwrite("linear_2.jpg", matDst2);

3、双三次：由相邻的4*4像素计算得出，公式类似于双线性

  1. 	int iscale_x = cv::saturate_cast<int>(scale_x);
  2. 	int iscale_y = cv::saturate_cast<int>(scale_y);
  3.  
  4. 	for (int j = 0; j < matDst1.rows; ++j)
  5. 	{
  6. 		float fy = (float)((j + 0.5) * scale_y - 0.5);
  7. 		int sy = cvFloor(fy);
  8. 		fy -= sy;
  9. 		sy = std::min(sy, matSrc.rows - 3);
  10. 		sy = std::max(1, sy);
  11.  
  12. 		const float A = -0.75f;
  13.  
  14. 		float coeffsY[4];
  15. 		coeffsY[0] = ((A*(fy + 1) - 5*A)*(fy + 1) + 8*A)*(fy + 1) - 4*A;
  16. 		coeffsY[1] = ((A + 2)*fy - (A + 3))*fy*fy + 1;
  17. 		coeffsY[2] = ((A + 2)*(1 - fy) - (A + 3))*(1 - fy)*(1 - fy) + 1;
  18. 		coeffsY[3] = 1.f - coeffsY[0] - coeffsY[1] - coeffsY[2];
  19.  
  20. 		short cbufY[4];
  21. 		cbufY[0] = cv::saturate_cast<short>(coeffsY[0] * 2048);
  22. 		cbufY[1] = cv::saturate_cast<short>(coeffsY[1] * 2048);
  23. 		cbufY[2] = cv::saturate_cast<short>(coeffsY[2] * 2048);
  24. 		cbufY[3] = cv::saturate_cast<short>(coeffsY[3] * 2048);
  25.  
  26. 		for (int i = 0; i < matDst1.cols; ++i)
  27. 		{
  28. 			float fx = (float)((i + 0.5) * scale_x - 0.5);
  29. 			int sx = cvFloor(fx);
  30. 			fx -= sx;
  31.  
  32. 			if (sx < 1) {
  33. 				fx = 0, sx = 1;
  34. 			}
  35. 			if (sx >= matSrc.cols - 3) {
  36. 				fx = 0, sx = matSrc.cols - 3;
  37. 			}
  38.  
  39. 			float coeffsX[4];
  40. 			coeffsX[0] = ((A*(fx + 1) - 5*A)*(fx + 1) + 8*A)*(fx + 1) - 4*A;
  41. 			coeffsX[1] = ((A + 2)*fx - (A + 3))*fx*fx + 1;
  42. 			coeffsX[2] = ((A + 2)*(1 - fx) - (A + 3))*(1 - fx)*(1 - fx) + 1;
  43. 			coeffsX[3] = 1.f - coeffsX[0] - coeffsX[1] - coeffsX[2];
  44.  
  45. 			short cbufX[4];
  46. 			cbufX[0] = cv::saturate_cast<short>(coeffsX[0] * 2048);
  47. 			cbufX[1] = cv::saturate_cast<short>(coeffsX[1] * 2048);
  48. 			cbufX[2] = cv::saturate_cast<short>(coeffsX[2] * 2048);
  49. 			cbufX[3] = cv::saturate_cast<short>(coeffsX[3] * 2048);
  50.  
  51. 			for (int k = 0; k < matSrc.channels(); ++k)
  52. 			{
  53. 				matDst1.at(j, i)[k] = abs((matSrc.at(sy-1, sx-1)[k] * cbufX[0] * cbufY[0] + matSrc.at(sy, sx-1)[k] * cbufX[0] * cbufY[1] +
  54. 					matSrc.at(sy+1, sx-1)[k] * cbufX[0] * cbufY[2] + matSrc.at(sy+2, sx-1)[k] * cbufX[0] * cbufY[3] +
  55. 					matSrc.at(sy-1, sx)[k] * cbufX[1] * cbufY[0] + matSrc.at(sy, sx)[k] * cbufX[1] * cbufY[1] +
  56. 					matSrc.at(sy+1, sx)[k] * cbufX[1] * cbufY[2] + matSrc.at(sy+2, sx)[k] * cbufX[1] * cbufY[3] +
  57. 					matSrc.at(sy-1, sx+1)[k] * cbufX[2] * cbufY[0] + matSrc.at(sy, sx+1)[k] * cbufX[2] * cbufY[1] +
  58. 					matSrc.at(sy+1, sx+1)[k] * cbufX[2] * cbufY[2] + matSrc.at(sy+2, sx+1)[k] * cbufX[2] * cbufY[3] +
  59. 					matSrc.at(sy-1, sx+2)[k] * cbufX[3] * cbufY[0] + matSrc.at(sy, sx+2)[k] * cbufX[3] * cbufY[1] +
  60. 					matSrc.at(sy+1, sx+2)[k] * cbufX[3] * cbufY[2] + matSrc.at(sy+2, sx+2)[k] * cbufX[3] * cbufY[3] ) >> 22);
  61. 			}
  62. 		}
  63. 	}
  64. 	cv::imwrite("cubic_1.jpg", matDst1);
  65.  
  66. 	cv::resize(matSrc, matDst2, matDst1.size(), 0, 0, 2);
  67. 	cv::imwrite("cubic_2.jpg", matDst2);

4、基于像素区域关系：共分三种情况，图像放大时类似于双线性插值，图像缩小(x轴、y轴同时缩小)又分两种情况，此情况下可以避免波纹出现。

  1. #ifdef _MSC_VER
  2. 	cv::resize(matSrc, matDst2, matDst1.size(), 0, 0, 3);
  3. 	cv::imwrite("E:/GitCode/OpenCV_Test/test_images/area_2.jpg", matDst2);
  4. #else
  5. 	cv::resize(matSrc, matDst2, matDst1.size(), 0, 0, 3);
  6. 	cv::imwrite("area_2.jpg", matDst2);
  7. #endif
  8.  
  9. 	fprintf(stdout, "==== start area ====
");
  10. 	double inv_scale_x = 1. / scale_x;
  11. 	double inv_scale_y = 1. / scale_y;
  12. 	int iscale_x = cv::saturate_cast<int>(scale_x);
  13. 	int iscale_y = cv::saturate_cast<int>(scale_y);
  14. 	bool is_area_fast = std::abs(scale_x - iscale_x) < DBL_EPSILON && std::abs(scale_y - iscale_y) < DBL_EPSILON;
  15.  
  16. 	if (scale_x >= 1 && scale_y >= 1)  { // zoom out
  17. 		if (is_area_fast)  { // integer multiples
  18. 			for (int j = 0; j < matDst1.rows; ++j) {
  19. 				int sy = std::min(cvFloor(j * scale_y), matSrc.rows - 1);
  20.  
  21. 				for (int i = 0; i < matDst1.cols; ++i) {
  22. 					int sx = std::min(cvFloor(i * scale_x), matSrc.cols -1);
  23.  
  24. 					matDst1.at(j, i) = matSrc.at(sy, sx);
  25. 				}
  26. 			}
  27. #ifdef _MSC_VER
  28. 			cv::imwrite("E:/GitCode/OpenCV_Test/test_images/area_1.jpg", matDst1);
  29. #else
  30. 			cv::imwrite("area_1.jpg", matDst1);
  31. #endif
  32. 			return 0;
  33. 		}
  34.  
  35. 		for (int j = 0; j < matDst1.rows; ++j) {
  36. 			double fsy1 = j * scale_y;
  37. 			double fsy2 = fsy1 + scale_y;
  38. 			double cellHeight = cv::min(scale_y, matSrc.rows - fsy1);
  39.  
  40. 			int sy1 = cvCeil(fsy1), sy2 = cvFloor(fsy2);
  41.  
  42. 			sy2 = std::min(sy2, matSrc.rows - 2);
  43. 			sy1 = std::min(sy1, sy2);
  44.  
  45. 			float cbufy[2];
  46. 			cbufy[0] = (float)((sy1 - fsy1) / cellHeight);
  47. 			cbufy[1] = (float)(std::min(std::min(fsy2 - sy2, 1.), cellHeight) / cellHeight);
  48.  
  49. 			for (int i = 0; i < matDst1.cols; ++i) {
  50. 				double fsx1 = i * scale_x;
  51. 				double fsx2 = fsx1 + scale_x;
  52. 				double cellWidth = std::min(scale_x, matSrc.cols - fsx1);
  53.  
  54. 				int sx1 = cvCeil(fsx1), sx2 = cvFloor(fsx2);
  55.  
  56. 				sx2 = std::min(sx2, matSrc.cols - 2);
  57. 				sx1 = std::min(sx1, sx2);
  58.  
  59. 				float cbufx[2];
  60. 				cbufx[0] = (float)((sx1 - fsx1) / cellWidth);
  61. 				cbufx[1] = (float)(std::min(std::min(fsx2 - sx2, 1.), cellWidth) / cellWidth);
  62.  
  63. 				for (int k = 0; k < matSrc.channels(); ++k) {
  64. 					matDst1.at(j, i)[k] = (uchar)(matSrc.at(sy1, sx1)[k] * cbufx[0] * cbufy[0] +
  65. 						matSrc.at(sy1 + 1, sx1)[k] * cbufx[0] * cbufy[1] +
  66. 						matSrc.at(sy1, sx1 + 1)[k] * cbufx[1] * cbufy[0] +
  67. 						matSrc.at(sy1 + 1, sx1 + 1)[k] * cbufx[1] * cbufy[1]);
  68. 				}
  69. 			}
  70. 		}
  71. #ifdef _MSC_VER
  72. 		cv::imwrite("E:/GitCode/OpenCV_Test/test_images/area_1.jpg", matDst1);
  73. #else
  74. 		cv::imwrite("area_1.jpg", matDst1);
  75. #endif
  76.  
  77. 		return 0;
  78. 	}
  79.  
  80. 	//zoom in,it is emulated using some variant of bilinear interpolation
  81. 	for (int j = 0; j < matDst1.rows; ++j) {
  82. 		int  sy = cvFloor(j * scale_y);
  83. 		float fy = (float)((j + 1) - (sy + 1) * inv_scale_y);
  84. 		fy = fy <= 0 ? 0.f : fy - cvFloor(fy);
  85. 		sy = std::min(sy, matSrc.rows - 2);
  86.  
  87. 		short cbufy[2];
  88. 		cbufy[0] = cv::saturate_cast<short>((1.f - fy) * 2048);
  89. 		cbufy[1] = 2048 - cbufy[0];
  90.  
  91. 		for (int i = 0; i < matDst1.cols; ++i) {
  92. 			int sx = cvFloor(i * scale_x);
  93. 			float fx = (float)((i + 1) - (sx + 1) * inv_scale_x);
  94. 			fx = fx < 0 ? 0.f : fx - cvFloor(fx);
  95.  
  96. 			if (sx < 0) {
  97. 				fx = 0, sx = 0;
  98. 			}
  99.  
  100. 			if (sx >= matSrc.cols - 1) {
  101. 				fx = 0, sx = matSrc.cols - 2;
  102. 			}
  103.  
  104. 			short cbufx[2];
  105. 			cbufx[0] = cv::saturate_cast<short>((1.f - fx) * 2048);
  106. 			cbufx[1] = 2048 - cbufx[0];
  107.  
  108. 			for (int k = 0; k < matSrc.channels(); ++k) {
  109. 				matDst1.at(j, i)[k] = (matSrc.at(sy, sx)[k] * cbufx[0] * cbufy[0] +
  110. 					matSrc.at(sy + 1, sx)[k] * cbufx[0] * cbufy[1] +
  111. 					matSrc.at(sy, sx + 1)[k] * cbufx[1] * cbufy[0] +
  112. 					matSrc.at(sy + 1, sx + 1)[k] * cbufx[1] * cbufy[1]) >> 22;
  113. 			}
  114. 		}
  115. 	}
  116. 	fprintf(stdout, "==== end area ====
");
  117.  
  118. #ifdef _MSC_VER
  119. 	cv::imwrite("E:/GitCode/OpenCV_Test/test_images/area_1.jpg", matDst1);
  120. #else
  121. 	cv::imwrite("area_1.jpg", matDst1);
  122. #endif

注：以上基于area进行图像缩小的代码有问题，具体实现代码可以参考http://iyenn.com/index/link?url=https://github.com/fengbingchun/OpenCV_Test/blob/master/src/fbc_cv/include/resize.hpp，用法如下：

  1. fbc::Mat3BGR src(matSrc.rows, matSrc.cols, matSrc.data);
  2. fbc::Mat3BGR dst(matDst1.rows, matDst1.cols, matDst1.data);
  3. fbc::resize(src, dst, 3);

5、兰索斯插值：由相邻的8*8像素计算得出，公式类似于双线性

  1. 	int iscale_x = cv::saturate_cast<int>(scale_x);
  2. 	int iscale_y = cv::saturate_cast<int>(scale_y);
  3.  
  4. 	for (int j = 0; j < matDst1.rows; ++j)
  5. 	{
  6. 		float fy = (float)((j + 0.5) * scale_y - 0.5);
  7. 		int sy = cvFloor(fy);
  8. 		fy -= sy;
  9. 		sy = std::min(sy, matSrc.rows - 5);
  10. 		sy = std::max(3, sy);
  11.  
  12. 		const double s45 = 0.70710678118654752440084436210485;
  13. 		const double cs[][2] = {{1, 0}, {-s45, -s45}, {0, 1}, {s45, -s45}, {-1, 0}, {s45, s45}, {0, -1}, {-s45, s45}};
  14. 		float coeffsY[8];
  15.  
  16. 		if (fy < FLT_EPSILON) {
  17. 			for (int t = 0; t < 8; t++)
  18. 				coeffsY[t] = 0;
  19. 			coeffsY[3] = 1;
  20. 		} else {
  21. 			float sum = 0;
  22. 			double y0 = -(fy + 3) * CV_PI * 0.25, s0 = sin(y0), c0 = cos(y0);
  23.  
  24. 			for (int t = 0; t < 8; ++t)
  25. 			{
  26. 				double dy = -(fy + 3 -t) * CV_PI * 0.25;
  27. 				coeffsY[t] = (float)((cs[t][0] * s0 + cs[t][1] * c0) / (dy * dy));
  28. 				sum += coeffsY[t];
  29. 			}
  30.  
  31. 			sum = 1.f / sum;
  32. 			for (int t = 0; t < 8; ++t)
  33. 				coeffsY[t] *= sum;
  34. 		}
  35.  
  36. 		short cbufY[8];
  37. 		cbufY[0] = cv::saturate_cast<short>(coeffsY[0] * 2048);
  38. 		cbufY[1] = cv::saturate_cast<short>(coeffsY[1] * 2048);
  39. 		cbufY[2] = cv::saturate_cast<short>(coeffsY[2] * 2048);
  40. 		cbufY[3] = cv::saturate_cast<short>(coeffsY[3] * 2048);
  41. 		cbufY[4] = cv::saturate_cast<short>(coeffsY[4] * 2048);
  42. 		cbufY[5] = cv::saturate_cast<short>(coeffsY[5] * 2048);
  43. 		cbufY[6] = cv::saturate_cast<short>(coeffsY[6] * 2048);
  44. 		cbufY[7] = cv::saturate_cast<short>(coeffsY[7] * 2048);
  45.  
  46. 		for (int i = 0; i < matDst1.cols; ++i)
  47. 		{
  48. 			float fx = (float)((i + 0.5) * scale_x - 0.5);
  49. 			int sx = cvFloor(fx);
  50. 			fx -= sx;
  51.  
  52. 			if (sx < 3) {
  53. 				fx = 0, sx = 3;
  54. 			}
  55. 			if (sx >= matSrc.cols - 5) {
  56. 				fx = 0, sx = matSrc.cols - 5;
  57. 			}
  58.  
  59. 			float coeffsX[8];
  60.  
  61. 			if (fx < FLT_EPSILON) {
  62. 				for ( int t = 0; t < 8; t++ )
  63. 					coeffsX[t] = 0;
  64. 				coeffsX[3] = 1;
  65. 			} else {
  66. 				float sum = 0;
  67. 				double x0 = -(fx + 3) * CV_PI * 0.25, s0 = sin(x0), c0 = cos(x0);
  68.  
  69. 				for (int t = 0; t < 8; ++t)
  70. 				{
  71. 					double dx = -(fx + 3 -t) * CV_PI * 0.25;
  72. 					coeffsX[t] = (float)((cs[t][0] * s0 + cs[t][1] * c0) / (dx * dx));
  73. 					sum += coeffsX[t];
  74. 				}
  75.  
  76. 				sum = 1.f / sum;
  77. 				for (int t = 0; t < 8; ++t)
  78. 					coeffsX[t] *= sum;
  79. 			}
  80.  
  81. 			short cbufX[8];
  82. 			cbufX[0] = cv::saturate_cast<short>(coeffsX[0] * 2048);
  83. 			cbufX[1] = cv::saturate_cast<short>(coeffsX[1] * 2048);
  84. 			cbufX[2] = cv::saturate_cast<short>(coeffsX[2] * 2048);
  85. 			cbufX[3] = cv::saturate_cast<short>(coeffsX[3] * 2048);
  86. 			cbufX[4] = cv::saturate_cast<short>(coeffsX[4] * 2048);
  87. 			cbufX[5] = cv::saturate_cast<short>(coeffsX[5] * 2048);
  88. 			cbufX[6] = cv::saturate_cast<short>(coeffsX[6] * 2048);
  89. 			cbufX[7] = cv::saturate_cast<short>(coeffsX[7] * 2048);
  90.  
  91. 			for (int k = 0; k < matSrc.channels(); ++k)
  92. 			{
  93. 				matDst1.at(j, i)[k] = abs((matSrc.at(sy-3, sx-3)[k] * cbufX[0] * cbufY[0] + matSrc.at(sy-2, sx-3)[k] * cbufX[0] * cbufY[1] +
  94. 					matSrc.at(sy-1, sx-3)[k] * cbufX[0] * cbufY[2] + matSrc.at(sy, sx-3)[k] * cbufX[0] * cbufY[3] +
  95. 					matSrc.at(sy+1, sx-3)[k] * cbufX[0] * cbufY[4] + matSrc.at(sy+2, sx-3)[k] * cbufX[0] * cbufY[5] +
  96. 					matSrc.at(sy+3, sx-3)[k] * cbufX[0] * cbufY[6] + matSrc.at(sy+4, sx-3)[k] * cbufX[0] * cbufY[7] +
  97.  
  98. 					matSrc.at(sy-3, sx-2)[k] * cbufX[1] * cbufY[0] + matSrc.at(sy-2, sx-2)[k] * cbufX[1] * cbufY[1] +
  99. 					matSrc.at(sy-1, sx-2)[k] * cbufX[1] * cbufY[2] + matSrc.at(sy, sx-2)[k] * cbufX[1] * cbufY[3] +
  100. 					matSrc.at(sy+1, sx-2)[k] * cbufX[1] * cbufY[4] + matSrc.at(sy+2, sx-2)[k] * cbufX[1] * cbufY[5] +
  101. 					matSrc.at(sy+3, sx-2)[k] * cbufX[1] * cbufY[6] + matSrc.at(sy+4, sx-2)[k] * cbufX[1] * cbufY[7] +
  102.  
  103. 					matSrc.at(sy-3, sx-1)[k] * cbufX[2] * cbufY[0] + matSrc.at(sy-2, sx-1)[k] * cbufX[2] * cbufY[1] +
  104. 					matSrc.at(sy-1, sx-1)[k] * cbufX[2] * cbufY[2] + matSrc.at(sy, sx-1)[k] * cbufX[2] * cbufY[3] +
  105. 					matSrc.at(sy+1, sx-1)[k] * cbufX[2] * cbufY[4] + matSrc.at(sy+2, sx-1)[k] * cbufX[2] * cbufY[5] +
  106. 					matSrc.at(sy+3, sx-1)[k] * cbufX[2] * cbufY[6] + matSrc.at(sy+4, sx-1)[k] * cbufX[2] * cbufY[7] +
  107.  
  108. 					matSrc.at(sy-3, sx)[k] * cbufX[3] * cbufY[0] + matSrc.at(sy-2, sx)[k] * cbufX[3] * cbufY[1] +
  109. 					matSrc.at(sy-1, sx)[k] * cbufX[3] * cbufY[2] + matSrc.at(sy, sx)[k] * cbufX[3] * cbufY[3] +
  110. 					matSrc.at(sy+1, sx)[k] * cbufX[3] * cbufY[4] + matSrc.at(sy+2, sx)[k] * cbufX[3] * cbufY[5] +
  111. 					matSrc.at(sy+3, sx)[k] * cbufX[3] * cbufY[6] + matSrc.at(sy+4, sx)[k] * cbufX[3] * cbufY[7] +
  112.  
  113. 					matSrc.at(sy-3, sx+1)[k] * cbufX[4] * cbufY[0] + matSrc.at(sy-2, sx+1)[k] * cbufX[4] * cbufY[1] +
  114. 					matSrc.at(sy-1, sx+1)[k] * cbufX[4] * cbufY[2] + matSrc.at(sy, sx+1)[k] * cbufX[4] * cbufY[3] +
  115. 					matSrc.at(sy+1, sx+1)[k] * cbufX[4] * cbufY[4] + matSrc.at(sy+2, sx+1)[k] * cbufX[4] * cbufY[5] +
  116. 					matSrc.at(sy+3, sx+1)[k] * cbufX[4] * cbufY[6] + matSrc.at(sy+4, sx+1)[k] * cbufX[4] * cbufY[7] +
  117.  
  118. 					matSrc.at(sy-3, sx+2)[k] * cbufX[5] * cbufY[0] + matSrc.at(sy-2, sx+2)[k] * cbufX[5] * cbufY[1] +
  119. 					matSrc.at(sy-1, sx+2)[k] * cbufX[5] * cbufY[2] + matSrc.at(sy, sx+2)[k] * cbufX[5] * cbufY[3] +
  120. 					matSrc.at(sy+1, sx+2)[k] * cbufX[5] * cbufY[4] + matSrc.at(sy+2, sx+2)[k] * cbufX[5] * cbufY[5] +
  121. 					matSrc.at(sy+3, sx+2)[k] * cbufX[5] * cbufY[6] + matSrc.at(sy+4, sx+2)[k] * cbufX[5] * cbufY[7] +
  122.  
  123. 					matSrc.at(sy-3, sx+3)[k] * cbufX[6] * cbufY[0] + matSrc.at(sy-2, sx+3)[k] * cbufX[6] * cbufY[1] +
  124. 					matSrc.at(sy-1, sx+3)[k] * cbufX[6] * cbufY[2] + matSrc.at(sy, sx+3)[k] * cbufX[6] * cbufY[3] +
  125. 					matSrc.at(sy+1, sx+3)[k] * cbufX[6] * cbufY[4] + matSrc.at(sy+2, sx+3)[k] * cbufX[6] * cbufY[5] +
  126. 					matSrc.at(sy+3, sx+3)[k] * cbufX[6] * cbufY[6] + matSrc.at(sy+4, sx+3)[k] * cbufX[6] * cbufY[7] +
  127.  
  128. 					matSrc.at(sy-3, sx+4)[k] * cbufX[7] * cbufY[0] + matSrc.at(sy-2, sx+4)[k] * cbufX[7] * cbufY[1] +
  129. 					matSrc.at(sy-1, sx+4)[k] * cbufX[7] * cbufY[2] + matSrc.at(sy, sx+4)[k] * cbufX[7] * cbufY[3] +
  130. 					matSrc.at(sy+1, sx+4)[k] * cbufX[7] * cbufY[4] + matSrc.at(sy+2, sx+4)[k] * cbufX[7] * cbufY[5] +
  131. 					matSrc.at(sy+3, sx+4)[k] * cbufX[7] * cbufY[6] + matSrc.at(sy+4, sx+4)[k] * cbufX[7] * cbufY[7] ) >> 22);// 4194304
  132. 			}
  133. 		}
  134. 	}
  135. 	cv::imwrite("Lanczos_1.jpg", matDst1);
  136.  
  137. 	cv::resize(matSrc, matDst2, matDst1.size(), 0, 0, 4);
  138. 	cv::imwrite("Lanczos_2.jpg", matDst2);

以上代码的实现结果与cv::resize函数相同，但是执行效率非常低，只是为了详细说明插值过程。OpenCV中默认采用C++ Concurrency进行优化加速，你也可以采用TBB、OpenMP等进行优化加速。

GitHub：http://iyenn.com/index/link?url=https://github.com/fengbingchun/OpenCV_Test/blob/master/demo/OpenCV_Test/test_opencv_funset.cpp

文章知识点与官方知识档案匹配，可进一步学习相关知识
算法技能树首页概览51048 人正在系统学习中
注：本文转载自blog.csdn.net的fengbingchun的文章"https://blog.csdn.net/fengbingchun/article/details/17335477"。版权归原作者所有，此博客不拥有其著作权，亦不承担相应法律责任。如有侵权，请联系我们删除。
复制链接
相关推荐
发表评论

评论记录：

未查询到任何数据！