Competitive Fuzzy-Classifier Image Edge Detection

The competitive fuzzy classifier operates on the set of four features extracted from the 3x3 neighborhood of each pixel. These features are the magnitudes of differences between that pixel and its neighboring pixels on four directions. They are input into the competitive fuzzy classifier inputs that connect to five fuzzy set membership functions that represent “white background” or one of the four different classes of “black edge”. Each pixel in the image is classified on its highest fuzzy truth and mapped to white or black accordingly. Class oriented magnitude competition is done before a “black edges” pixel is mapped to black to output fine lines. The paradigm is simple, computationally efficient, has low sensitivity to noise and is isotropic. The competitive fuzzy classifier yields thin black lines on a white background. The competitive fuzzy classifier thins the wide ridges around local maxima in difference magnitude down to edges that are only one pixel wide, and yields thin black lines on a white background. 2. Methodology For a center pixel p5 in a 3x3 neighborhood, we define four directions on its neighborhood: horizontal, vertical, and two diagonals. The gray-level difference magnitudes between p5 and its neighbors on these directions are designated by X1, X2, X3, X4, and calculated by X1= |P1-P5|+|P9-P5| X2= |P2-P5|+|P8-P5| X3= |P3-P5|+|P7-P5|