How are the edges of an image obtained with anisotropic diffusion?

Question

I already know how to implement the method and obtain a correctly diffused image, but in the papers related to the topic there is usually an edge map that's calculated after the diffusion is performed as an "edge detector scheme" (like in the original Perona and Malik paper) and compared to gaussian smoothing (canny detector) mentioning something related to scale-spaces. I'm making an implementation in MATLAB so I would like to know how is this edge map calculated as shown in the picture.

Answer 1

They compare the same idea of edge detection after some smoothing. They compare edge aware smoothing in the form of anisotropic diffusion vs. non edge aware method in the form of gaussian blur.