Combining image entropy with the Pulse Coupled Neural Network in edge detection

We propose a simple and effective approach for edge detection using the image entropy defined on pixel grayscale values instead of the histogram. A strictly bounded function of local image entropy is designed for identifying abrupt changes of image intensity across edges. Mathematical properties of this function are analyzed to validate its applicability in the edge detection task. Edge pixels are segmented using a Pulse Coupled Neural Network in which the connectivity prior of edge pixels is used. Experimental results demonstrate that our method can robustly detect edges in synthetic as well as natural images.