一种基于最大类间后验概率的Canny边缘检测算法

在分析了传统的Canny算法的基础上，用自适应滤波器代替原有的高斯滤波器，并利用交叉熵来度量目标和背景间的差异，结合贝叶斯判断理论，将这种类间差异性用原始图像中的像素点判决到目标和背景两类区域的后验概率之间的交叉熵的平均值来表示，通过最大化将像素点判决到不同区域的后验概率来获取最优的高低阈值。实验证明改进的算法具有很好的边缘检测效果。

Edge detection algorithm of Canny based on maximum between-class posterior probability

Based on the analysis of the traditional Canny algorithm, the adaptive filter took the place of the original Gaussian filter and made use of cross-entropy to measure the differences between the background and objectives. Combining Bayesian judgment theory, the average cross-entropy of posterior probability of the pixels of original image to objective and background areas presented differences between classes, and this paper maximized the posterior probability to judge pixels in which different regions to obtain the optimal level of the threshold. The experimental results show the improved algorithm has great edge detection effect.