Edge Detection Method Based on Hysteresis Connection and Prediction

Edge detection is a critical component of many vision systems, including object detectors and image segmentation algorithms. Based on the analysis of edge gradient changes and the defects of traditional edge detection algorithms, a novel edge detection method is proposed. Firstly, the edge set is initialized by gradient map binaryzation with a given threshold. Then the pixel adjacent to edge pixel will be joined to the edge set, if its gradient is higher than half of threshold or the gradient difference between this pixel and the adjacent strong edge is less than a given value. In order to go over the broken interval, this method predicts the edge extension direction using the local information of edge, and searches for discontinuous weak edges on this direction. The experimental results show that this method not only can extract complete edge segments, but also can detect the valid weak edge segment over the broken interval, and at the same time it can suppress the noise with the same gradient value.