An efficient edge detection algorithm using relaxation labeling technique

Edge detection plays an important role in computer vision tasks, and has received considerable attention in image processing literature. To detect edges correctly and precisely, contextual information is needed. How to use contextual information is a key issue. In this paper, we introduce an edge detection method that will use edge contextual information of the whole image efficiency. This new method tries to employ contextual information within a certain distance from the focus pixel at a time. This distance keeps increasing recursively until the edge feature of a pixel is uniquely defined. In this manner, we can minimize the need for contextual information. Experimental results are presented to characterize the performance of our new method in terms of better connectedness of edges and less distortion, and in terms of computational efficiency. A detailed comparison of our method with the context free zero-crossing edge operator that uses optimal exponential filter is discussed in this paper. Edge detection Probability Relaxation labeling Configuration diction&y Markov random field Recursive filtering