在红外图像的分割处理中，阈值的选取一直是研究的热点。针对传统Canny边缘检测算法需要人为设定 阈值的情况，从图像灰度的相似性和不连续性原理出发，提出了一种自适应图像边缘检测方法。该方法首先用中值滤噪代替Canny算法中 的高斯滤噪，以去除图像中的背景噪声。在此基础上，采用改进的最大类间方差法来优化Canny算法双阈值中高阈值的设定原则，以保 证双阈值的自适应性，再利用Canny多尺度空间边缘检测技术完成图像的分割处理。实验表明，与传统的单一算法相比，该算 法提高了信噪比，简化了运算过程，可达到比较理想的边缘检测和分割效果。

In infrared image segmentation processing, the selection of a threshold has been a hot issue. To solve the problem that a threshold should be set artificially in traditional Canny edge detection algorithms, an adaptive image edge detection method is proposed according to the similarity and non-continuity theory of image gray-scale. In the method, the Gaussian filtering is first displaced by the Median filtering so as to remove the backgroud noise in an image. Then, an improved Otsu method is used to optimize the setting principle of the higher threshold of a dual threshold in the Canny algorithm, so as to keep the dual threshold having its adaptability. Finally, the segmentation processing is completed by using the Canny multi-scale spacial edge detection technique. The experimental result shows that this improved algorithm has a higher signal to noise ratio and a simpler operation process than the traditional algorithm and can achieve ideal edge detection and segmentation.