EXC_BAD_ACCESS error OpenCV? - c++

Question

I'm trying write a code for image segmentation in OpenCV. As a part of the image processing, I'm trying to detect the edges of a test image using Sobel filter.
In order to find the magnitude of gradient on both dX and dY direction, I'm computing the euclidean distance of both the gradients. But when I run the code I get the above error. I do know that the above error occurs when I am trying "ACCESS" an unavailable location in memory, but I am sure I have defined all Mat in my code.
This is part of my code.
//Blur the raw image to remove noise
GaussianBlur(src, src, kernel, 2);
//Run sobel edge detector
Sobel(src, edgeX, src.depth(), 1, 0);
Sobel(src, edgeY, src.depth(), 0, 1);
edge = Mat::zeros(317,554,CV_8UC1);
for (int r = 0; r < edgeX.rows; r++)
{
for (int c = 0; c < edgeY.cols; c++)
{
edge.at<double>(r,c) = sqrt((edgeX.at<double>(r,c)*edgeX.at<double>(r,c)) + (edgeY.at<double>(r,c)*edgeY.at<double>(r,c)));
}
}
where:
src: the RGB test image
edgeX: sobel output with dX gradient
edgeY: sobel output with dY
edge: is the Mat with the euclidean distance.
I get the error at this line
edge.at<double>(r,c) = sqrt((edgeX.at<double>(r,c)*edgeX.at<double>(r,c)) + (edgeY.at<double>(r,c)*edgeY.at<double>(r,c)));
when trying to access edge.at<double>(316,395)
How do I debug this?
What am I doing wrong?

Answer 1

edge is a matrix of type CV_8UC1, which means a matrix of uchar, not of double.
You need to access it with at<uchar>:
edge.at<uchar>(r,c) = sqrt((edgeX.at<uchar>(r,c)*edgeX.at<uchar>(r,c)) + (edgeY.at<uchar>(r,c)*edgeY.at<uchar>(r,c)));
You can avoid this kind of problems using Mat_<Tp>, that allows also easier access without using the .at function:
Mat1b edge(317,554,uchar(0));
for (int r = 0; r < edgeX.rows; r++) {
for (int c = 0; c < edgeY.cols; c++) {
edge(r,c) = sqrt((edgeX(r,c)*edgeX(r,c)) + (edgeY(r,c)*edgeY(r,c)));
}
}
In this case, you can also use cv::magnitude which performs the same operation you're doing with your for loops (but it needs matrices of float):
Sobel(src, edgeX, CV_32F, 1, 0);
Sobel(src, edgeY, CV_32F, 0, 1);
Mat edge;
magnitude(edgeX, edgeY, edge);
// Convert to CV_8UC1
edge.convertTo(edge, CV_8UC1);