I implemented the Prewitt edge detector and compared my output with Matlab's in-built Prewitt edge detector and I noticed that my output gave thicker edges. What could be the reason?

input = imread('pic.png');
input = double(rgb2gray(input));
kernel_x = [-1, 0, 1; -1, 0, 1; -1, 0, 1];
kernel_y = [1, 1, 1; 0, 0, 0; -1, -1, -1];

[length, width] = size(input);
new = input;
for i = 1:length - 2
  for j = 1:width - 2
    Gx = sum(sum(kernel_x.*input(i:i 2, j:j 2)));
    Gy = sum(sum(kernel_y.*input(i:i 2, j:j 2)));
    new(i 1, j 1) = sqrt(Gx.^2   Gy.^2);
  end
end
new = uint8(new);

% binarizing image and setting threshold
edge = imbinarize(edge, 100); % final output from implementation

The in-built function I used is edge(input, 'Prewitt') Output from my implementation vs the in-built operator:

What could be the reason for this?

I tried changing the Threshold value also but still no luck.

CodePudding user response：

It is not clearly stated in the documentation, but for the Sobel, Prewitt and Roberts methods, edge applies a thinning. There is an optional input argument 'nothinning', which skips the thinning step. Unfortunately this is the only bit in the documentation that indicates that this step is happening.

You can apply a thinning if you have the Image Processing Toolbox using bwmorph(bw,'thin').