Research On Multi-modality Image Matching Technology Based On Significant Contour Information

Image matching technology is widely applied in these fields such as visual navigation, target searching and medical fields. There are a lot of researches on homologous image matching, but for multi-sensor image matching there are plenty of problems remain to be resolved. This article studies mainly on fields like target contour extraction, the robustness of contour character, angle correction, etc.First, in view of the problem that the existing contour extraction algorithm can't come to a satisfactory result being used under complex background, this article proposed an improved meanshift clustering algorithm which using entropy image information as prior knowledge to adaptively adjust the size of sampling window. In addition, by dynamically changing the kernel function and the parameter setting, it improves the real-time performance. Experimental results show that the algorithm can extract precise contour in a complex environment with about 40% reduction in computation time.Second, in order to improve the real-time and accuracy of the subsequent algorithm, this paper puts forward a kind of artificial target contour detection algorithm. This filtered algorithm for contour keeps only those contours of potential artificial target which contain adequate information and a neat shape.Then, to solve the problem that present matching algorithm based on contour information can't ideally complete the matching between partial and deformed contours, this paper puts forward a new local contour character description method which has a good robustness to deformation.Finally, in view of the problem that actual matching using only local contour information will produce lots of mismatch, this article uses the PNP algorithm to adjust all the matching points into the same angle of view. Then we can judge whether the matching is correct by comparing the coordinates of all the adjusted matching points.