CODEWITHSUNDEEP
matlabimage-processingfiltercomputer-visionconvolution
jlcv
jlcv

Reputation: 1808

Flipped conv2 output when trying to apply a Sobel filter

I have the following image that I am convolving with a Sobel filter:

enter image description here

At row 157, the color begins to change and we see that the intensity increases. Note in my MATLAB code, I converted the image to grayscale. This behavior can be seen below:

enter image description here

I then apply the following Sobel Filter:

Gy = [-1 -2 -1; 0 0 0; +1 +2 +1];
A = conv2(im, Gy);

I expect the output to be a matrix with positive numbers since the intensity increases. However, the output shows a matrix with negative numbers. Does anyone know why?

enter image description here

Upvotes: 3

Views: 1252

Answers (1)

rayryeng
rayryeng

Reputation: 104503

conv2 rotates the kernel / filter mask by 180 degrees before performing filtering. This is the definition of convolution (a.k.a. you're using conv2 - 2D convolution). The positive changes you're expecting as the result is from the process of performing correlation instead. As such, either use filter2, or rotate the kernel by 180 degrees by doing Gy = Gy(end:-1:1, end:-1:1); and then call conv2 after so that inside conv2 it will rotate the kernel again so that you're performing correlation instead.

As such:

Gy = [-1 -2 -1; 0 0 0; +1 +2 +1];
Gy = Gy(end:-1:1, end:-1:1); %// Added
A = conv2(im, Gy);

Similarly, if you have access to the image processing toolbox, you may use imfilter as already suggested by another answer and the default behaviour is to use correlation instead of convolution. You can override this behaviour by specifying the 'conv' flag if you expressly want to perform convolution instead.

Upvotes: 3

Related Questions