Question

我试图在我的图像区域上进行卷积，只需一个像素，就更准确了。我还希望在计算此ROI的边界时，此卷积使用ROI之外的值。为了测试这个，我写了下面的代码。

float kernelData[25] = {0, 0, 0, 0, 0,
                        0, 0, 1, 0, 0,
                        0, 0, 1, 0, 0,
                        0, 0, 0, 0, 0,
                        0, 0, 0, 0, 0};

float imgData[9*9] = {0,  0,  0,   0,   0,    0,  0,   0,  0,
                      0,  0,  0,   0,   0,    0,  0,   0,  0,
                      0,  0,  0,   0,   0,    0,  0,   0,  0,
                      0,  0,  0,   2, 256,  128,  0,   0,  0,
                      0,  0,  0,   4,   1,   64,  0,   0,  0,
                      0,  0,  0,   8,  16,   32,  0,   0,  0,
                      0,  0,  0,   0,   0,    0,  0,   0,  0,
                      0,  0,  0,   0,   0,    0,  0,   0,  0,
                      0,  0,  0,   0,   0,    0,  0,   0,  0};

Mat img = Mat(9, 9, CV_32FC1, (void*)imgData);
Mat kernel = Mat(5, 5, CV_32FC1, (void*)kernelData);
Mat window = Mat(img, Rect2i(Point2i(4, 4), Size(1, 1)));

Mat convolved1 = Mat(1, 1, CV_32FC1);
Mat convolved2 = Mat(1, 1, CV_32FC1);
Mat convolved3 = Mat(1, 1, CV_32FC1);
Mat convolved4 = Mat(1, 1, CV_32FC1);
Mat convolved5 = Mat(1, 1, CV_32FC1);

cv::filter2D(window, convolved1, CV_32F, kernel, Point(-1, -1), 0, BORDER_CONSTANT);
cv::filter2D(window, convolved2, CV_32F, kernel, Point(-1, -1), 0, BORDER_REPLICATE);
cv::filter2D(window, convolved3, CV_32F, kernel, Point(-1, -1), 0, BORDER_REFLECT);
cv::filter2D(window, convolved4, CV_32F, kernel, Point(-1, -1), 0, BORDER_REFLECT_101);
cv::filter2D(window, convolved5, CV_32F, kernel, Point(-1, -1), 0, BORDER_TRANSPARENT);

cout << img << endl;
cout << kernel << endl;
cout << window << endl;
cout << convolved1 << endl;
cout << convolved2 << endl;
cout << convolved3 << endl;
cout << convolved4 << endl;
cout << convolved5 << endl;

此代码的执行显示了以下结果：

[0, 0, 0, 0, 0, 0, 0, 0, 0;
 0, 0, 0, 0, 0, 0, 0, 0, 0;
 0, 0, 0, 0, 0, 0, 0, 0, 0;
 0, 0, 0, 2, 256, 128, 0, 0, 0;
 0, 0, 0, 4, 1, 64, 0, 0, 0;
 0, 0, 0, 8, 16, 32, 0, 0, 0;
 0, 0, 0, 0, 0, 0, 0, 0, 0;    
 0, 0, 0, 0, 0, 0, 0, 0, 0;    
 0, 0, 0, 0, 0, 0, 0, 0, 0]
[0, 0, 0, 0, 0;
 0, 0, 1, 0, 0;
 0, 0, 1, 0, 0;
 0, 0, 0, 0, 0;
 0, 0, 0, 0, 0]
[1]
[5]
[5]
[5]
[5]
[5]

我测试了所有边框类型，但没有一个得到正确的结果，257。我也尝试修改内核，改变1和0的位置，我得到的结果总是错的。

有趣的是，这个相同的代码在我的笔记本上正确运行，但在我的桌面上却没有。

为什么会这样？

OpenCV对ROI的卷积

0 个答案: