Abstract: In classical image edge detection algorithms, gradient images are obtained by gradient operator convoluting, and contours are traced according to the gradient changing and gradient threshold. But the uneven changing of local gradient in image leads to not accurate edge information obtained through uniform threshold segmentation. To solve these problems, in this paper an adapted threshold method is introduced based on the expectation of local block in image. First, we get the image gradient matrix with the help of Sobel operator, then segment the matrix into a plurality of sub regions, and calculate the local expectations of each sub region as the threshold to each of them and filter out the edges of each sub regions finally. The experiment results show that the proposed method can enhance the recognition of primary object edge detection in image effectively, and the edge information obtained is accurate.

Key words: edge detection, Sobel operator, local expectation, threshold segmentation