一种改进的基于不变矩的图像匹配算法

An Improved Image Matching Algorithm Based on Moment Invariant

通过对不变矩理论的分析,提出以局部图像的边缘不变矩特征及其灰度信息的统计特征相结合的两阶段匹配算法。首先利用不变矩的平移、旋转及伸缩的不变性,计算图像边缘不变矩,将其作为匹配特征,在匹配结果中取前N个较好的匹配结果作为侯选目标;再根据契比雪夫不等式确定候选目标的置信度;最后计算各个候选目标的直方图包含的统计特性进行最终目标的定位。实验结果表明该算法克服了传统图像匹配方法搜索目标时存在的置信度问题,通过与基于不变矩特征直接匹配方法和基于灰度相关的匹配方法比较可见,在性能上明显优于这两个方法,并得到较理想的匹配结果。

This paper proposes a two--stage template--based matching method for object tracking under complex background. This method combines features of moment invariants and statistics of gray level histogram in local region by means of analyzing the theory of moment invariant. First of all, this algorithm that utilizes features of translation, rotation and scale of moment invariant as matching computes moment invariant of edges of image and obtains frist N better results as candidate targets in the first matching stage. Secondly, the algorithm based on Chebyshev inequation, which determines confidence level of candidate targets, decides the best target through the gray level histogram characteristics of candidate targets whose confidence level are higher. Those experimental results present that this algorithm addresses the problem of confidence level, which generally exists in traditional matching methods. The performance of this algorithm is superior to those of both directly matching method based on moment invariant and gray--level correlation matching method. Finally, a good matching result is obtained.