Edge detection has been an important domain in comprehension of image and computer vision, Its quality determines the result of subsequent analysis. Thus it＇s an important goal for people to find a kind of method that is insensitive to noise, precisely locates true edges and excludes false edges. In theory Canny operator is optimal in continuous domain for edge detection, but not optimal in discrete case, which may lead to inaccurate result. In order to solve the problem above, a good edge detection algorithm in discrete field is proposed. The smooth operator is inferred based on the Optimal Discrete Filter＇s theory, which can reduce the influence of noise, false edges and image error. The experimental result shows that the proposed method can achieve multi-scale edge detection and high SNR, satisfy the requirement for computer recognition. So it is a good edge detection method.