Abstract: An edge detection method is proposed combining image discrete information entropy with self-organizing map (SOM). A threshold is chosen from some different information entropies to segment the smooth region from the region where the gray level abruptly changes so as to reduce computation. Then, the gray level images are transformed into the ideal binary pattern of pixels. Six types of edge and six prototype vectors are defined, among them the latter are taken as inputs into SOM to classify the edge types and then obtain edge images from which the speckled edges are removed to acquire ideal edge images. Experimental results showed that the edge images gained by the method proposed are better than those by other edge detection methods.