CS50 模糊滤镜在小测试用例中给出了不正确的输出

Question

我正在为 CS50（更舒适的版本）中的过滤任务编写均值模糊过滤器。当我用我的一些照片测试它时，它似乎工作正常。但是，当我使用 CS50 测试用例对其进行测试时，它给出了小矩阵的错误结果。下面是我的代码

#define KERNEL_HEIGHT 3
#define KERNEL_WIDTH 3
#define KERNEL_SIZE 9
#define CORNER 4
#define EDGE 6
void blur(int height, int width, RGBTRIPLE image[height][width])
{
    // height and width are the height and width of the input image.
    RGBTRIPLE(*extension)[width + 2] = calloc(height + 2, (width + 2) * sizeof(RGBTRIPLE)); // I will handle edge and corner cases by creating a new 2d array with 0s around 
    RGBTRIPLE(*newImage)[width] = calloc(height, width * sizeof(RGBTRIPLE)); //output 
    int cof = 0, redSum = 0, greenSum = 0, blueSum = 0;
    int kernel[3][3] = {{1, 1, 1}, {1, 1, 1}, {1, 1, 1}}; // Kernel for mean blur
    if (newImage == NULL || extension == NULL)
    {
        fprintf(stderr, "Not enough memory to create a new image.\n");
        return;
    }
    for(int i = 0; i < height; ++i)
    {
        memcpy(extension[i + 1], image[i], sizeof(image[i])); // Copy values from original images into extension, note that I intentionally copy at the i + 1 position.
    }
    for(int i = 1; i < height + 1; ++i)
    {
        for(int j = 1; j < width + 1; ++j)
        {
            redSum = 0; greenSum = 0; blueSum = 0; //Reset red, green, blue sum values.
            for(int k = 0; k < KERNEL_HEIGHT; ++k)
            {
                for(int t = 0; t < KERNEL_WIDTH; ++t) 
                {
                    cof = kernel[k][t]; // I will multiply each value in the kernel with the pixel value and its neighbors.
                    redSum += cof * extension[i - 1 + k][j - 1 + t].rgbtRed;
                    greenSum += cof * extension[i - 1 + k][j - 1 + t].rgbtGreen;
                    blueSum += cof * extension[i - 1 + k][j - 1 + t].rgbtBlue;
                }
            }
            if(i == 1 || i == height) // Detecting edge cases and corner
            {
                if(j == 1 || j == width) // Corner
                {
                    newImage[i - 1][j - 1].rgbtRed = (int) round((float) redSum / (float) CORNER);
                    newImage[i - 1][j - 1].rgbtGreen = (int) round((float) greenSum / (float) CORNER);
                    newImage[i - 1][j - 1].rgbtBlue = (int) round((float) blueSum / (float) CORNER);
                }
                else // Edges
                {
                    newImage[i - 1][j - 1].rgbtRed = (int) round((float) redSum / (float) EDGE);
                    newImage[i - 1][j - 1].rgbtGreen = (int) round((float) greenSum / (float) EDGE);
                    newImage[i - 1][j - 1].rgbtBlue = (int) round((float) blueSum / (float) EDGE);
                }
            }
            else //Normal cases
            {
                newImage[i - 1][j - 1].rgbtRed = (int) round((float) redSum / (float) KERNEL_SIZE);
                newImage[i - 1][j - 1].rgbtGreen = (int) round((float) greenSum / (float) KERNEL_SIZE);
                newImage[i - 1][j - 1].rgbtBlue = (int) round((float) blueSum / (float) KERNEL_SIZE);
            }
        }
    }
    for(int i = 0; i < height; ++i)
    {
        memcpy(image[i], newImage[i], sizeof(newImage[i])); // Copy back the results into my original image.
    }
    free(extension);
    free(newImage);
    return;
}

这是测试用例之一：
使用样本 3x3 图像进行测试（过滤中间像素）
第一行：(10, 20, 30), (40, 50, 60), (70, 80, 90)
第二行：(110, 130, 140), (120, 140, 150), (130, 150, 160)
第三行：(200, 210, 220), (220, 230, 240), (240, 250, 255)

预期输出：127 140 149
实际输出：91 100 106

几天来我一直在尝试修复此功能（gg 搜索、阅读文档），但仍然没有结果。我认为这与我如何为扩展矩阵分配内存以及如何遍历它有关。我对动态内存分配还很陌生，所以也许这就是我出错的部分。

Answer 1

我讨厌这样做，当我有更多时间时会重新访问但是...以下内容尚未经过测试且有意不完整 - 仅用作风格指南。

我将post BlurPixel() 的大部分内容向您展示如何以合乎逻辑的方式处理每个像素。请注意，我们没有疯狂的索引，并且每次写入都由 if 语句进行边界检查。

      int r=0, g=0, b=0;  // Sums accumulators
      int numElements = 0; // Counts number of items accumulated.

      // Loop from 1 less to 1 more than the current X and Y position.
      for(int xOffset=-1; xOffset<=1; xOffset++)
      {
        for(int yOffset=-1; yOffset<=1; yOffset++)
        {
          const int newX = x + xOffset;
          const int newY = y + yOffset;

          // Bounds check... 
          if(newX >= 0 && newX < width &&
             newY >= 0 && newY < height) 
          {
            // In range, add to accumulators and increment element counter.
            r += image[newY][newX].red;
            g += image[newY][newX].green;
            b += image[newY][newX].blue;
            numElements++;
          }
        }
      }  

      // Assign the rounded values to the new structure.  Could do this with integers
      // if we added numElements/2 to list before divide.
      newImage[j][i].red = (int)round((float) r / numElements);
      newImage[j][i].green = (int)round((float) g / numElements);
      newImage[j][i].blue = (int)round((float) b / numElements);

CS50 模糊滤镜在小测试用例中给出了不正确的输出

CS50 Blur filter gave incorrect outputs with small test cases

c

image-processing

convolution

cs50