一种图像特征点迭代修正匹配算法

基于特征点的图像匹配被广泛应用于图像配准、目标识别与跟踪领域, 目前, 两阶段匹配(即先粗匹配, 后精匹配)是最常用的方法, 然而, 两阶段匹配存在两方面的问题, 一方面, 粗匹配阶段对精匹配阶段的影响是不可逆的, 即粗匹配的效果决定了精匹配的最优精度; 另一方面, 精匹配得到的后验知识没能反馈给粗匹配阶段, 以修正粗匹配结果. 为此, 提出一种基于迭代修正的图像特征点匹配算法, 该算法将精匹配得到的后验知识反馈给粗匹配阶段, 从而修正粗匹配结果, 使得粗匹配阶段得到更多的正确匹配对, 减少漏匹配特征点对, 这样经过多次迭代, 能够得到更多的正确匹配特征点对. 实验表明, 提出的算法比经典的两阶段匹配方法能够提取更多的正确匹配特征点对, 减少了漏匹配, 并提升了复杂图像匹配的稳定性.

Image Feature Point Matching Algorithm Based on Iterative Correction

Based on feature point, image matching has been widely applied in image registration, object recognition and tracking field. Now, two phase feature point matching (i.e., first coarse matching, then precise matching) is the most commonly used method. However, the two phase matching exists two issues, on the one hand, the impact of coarse matching for precise matching is irreversible, that is, the results of coarse matching will determine the optimal precision of precise matching. On the other hand, the post knowledge which can be obtained from precise matching cannot be regarded as feedback information to coarse matching, which can revise mismatching. Hence, the paper proposes a new feature point matching algorithm which is based on iterative correction. In the algorithm, post knowledge of precise matching is regarded as feedback information to coarse matching. The coarse matching can obtain more correct matching pairs and decrease missing correct matching pairs. After much iterations, better matching can be obtained. Experiments show that the proposed algorithm can extract more matching pairs than traditional two phase method and improves the matching stability.