Abstract: A new Harris-SIFT matching method based on edge information is proposed to solve the visible and infrared image matching problem. Firstly, in preprocessing procedure, the Retinex edge enhancement method is performed on the infrared image, and then the bilateral filtering processing is performed on the visible and the processed infrared image separately in order to enhance the similarity of the edge structure of two types of images. Secondly, Sobel operator is used to extract edge information, and an algorithm based on the Harris-SIFT method is adopted to detect the feature points in a edge detection image and build corresponding feature descriptors. Next, a method using the ratio of the closest neighbor and second closest neighbor is included in the rough matching process and RANSAC algorithm in the precise matching. At last, matching accuracy is further improved by means of a quick checking method that verifies the consistency of structural characteristics of the extended neighborhood around the matching pairs in edge detection images. Final results of experiments with different images demonstrate the effectiveness and value of the proposed method for improving the accuracy of visible and infrared image matching.

Key words: image matching, edge extraction, Harris-SIFT, neighbor verification

摘要： 针对可见光图像和红外图像的匹配问题，提出一种基于边缘信息的Harris-SIFT匹配算法。匹配过程中，对红外图像进行Retinex边缘增强，再分别对可见光图像和处理后的红外图像进行双边滤波预处理，以增强两类图像边缘结构的相似度。采用Sobel算子提取图像边缘信息，并基于Harris-SIFT算法对边缘信息进行特征点检测和描述，再分别通过最近邻和次近邻比值法，以及RANSAC算法完成粗匹配和精匹配。在精匹配的基础上，通过快速校验边缘图像中匹配点对在扩展邻域内的结构特征一致性，检测并筛除部分误匹配点对，从而进一步提高匹配精度。通过采用多幅图像进行实验，处理结果表明，该算法能够有效改善可见光图像与红外图像的匹配效果。

关键词: 图像匹配, 边缘提取, Harris-SIFT, 邻域校验