边缘是图像最基本的特征，是图像分割的第一步。经典的边缘检测方法如：Roberts，Sobel，Prewitt，Kirsch，Laplace等方法，基本都是对原始图像中象素的小邻域构造边缘检测算子，进行行一阶微分或二阶微分运算，求得梯度最大值或二阶导数的过零点，最后选取适当的阀值提取边界。由于这些算法涉及梯度的运算，因此均存在对噪声敏感、计算量大等缺点。在实践中，发现SUSAN算法只基于对周边象素的灰

-Edge of the basic characteristics of the image, is the first step in image segmentation. Classical edge detection methods such as: Roberts, Sobel, Prewitt, Kirsch, and, Laplace, and other methods are basically small neighborhood structure of the original image pixel edge detection operator to carry the line of the first derivative or second-order differential operator, seeking gradient maximum or second derivative zero-crossing, and finally select the appropriate threshold to extract boundary. Because these algorithms involve the computation of the gradient, and therefore are sensitive to noise, and calculate the volume shortcomings. In practice, the SUSAN algorithm based on the surrounding pixel gray