Question

I’ve just tested Sobel using C api and C++ api. But why is it different? All the parameters I used are the same.

Output – using C API

Output – using C++ API

Edited

C API :

/// Generate grad_x
grad_x = cvCreateImage(cvGetSize(grayImg), IPL_DEPTH_16S, 1);
abs_grad_x = cvCreateImage(cvGetSize(grayImg), 8, 1);

/// Gradient X
cvSobel(grayImg, grad_x, 1, 0, 3);
cvConvertScaleAbs(grad_x, abs_grad_x);
cvThreshold(abs_grad_x, abs_grad_x, 0, 255, CV_THRESH_BINARY|CV_THRESH_OTSU);

C++ API :

cv::Mat img_sobel;
cv::Sobel(img_gray, img_sobel, CV_8U, 1, 0, 3, 1, 0, BORDER_DEFAULT);

Mat img_threshold;
threshold(img_sobel, img_threshold, 0, 255, CV_THRESH_OTSU+CV_THRESH_BINARY);

Answer 1

There is only 1 reason for different results. Data-Type!

In the C version, you are creating grad_x with IPL_DEPTH_16S depth. So each pixel has short data type. This increases the precision of the results which you get when calling the cvSobel function. cvSobel is able to accommodate a wider range of values (-32768 to 32767) in the result grad_x.

In the C++ version, you are not initializing the matrix and specifying the destination type CV_8U. The function cv::Sobel internally creates destination matrix of type CV_8U, calculates the results and then clamps them to the range of destination data type, i.e from 0 to 255. So all the negative values become 0.

To get the same results in C version, change the IPL_DEPTH_16S to IPL_DEPTH_8U.