Abstract: According to the principle of the image edge formation, the image edges consist of the triangle edge and the ramp edge. Based on orientation average, this paper proposes a new image edge detection algorithm which segments the neighborhood of the center pixel into two semicircles along different direction and then calculates the difference of the mean values of the two semicircles. With the directions of maximum difference and the minimum difference, an edge magnitude response function is designed to judge the edge type of the pixel, the triangle edge or the ramp edge. For inhibiting the effects of the noise, this method adopts Gauss-smooth pre-filter which inhibits the noise in images effectively. The edge-detection fitness function combining omission ratio is designed. By utilizing the fitness function, the algorithm analyzes the relationship of the size of the smoothing filter and the radius of the neighborhood of the center pixel. Experimental results show that the algorithm has good precision, and to a certain extent, inhibits the effects of noise on edge detection.