Stepped-down coefficient values associated with Hopfield nets improve optimal edge detection

We (1999) have shown that the periodic reduction in the value of the coefficients of the non-quadratic energy functionals associated with Hopfield-type neural networks reaches stable state faster (with less number of iterations required). Here, we apply such an algorithm to the edge detection problem and compare its performance with those of Sobel (1970), Johnson (1990), Laplacian-of-Gaussian and Canny's (1986) edge detection algorithms, both quantitatively and visually. The test images used include both noisy and noise-free artificial checkerboard and circle images and four different men-made and natural, textured and nontextured publicly available images.