Multi-temporal Image Matching Technology

Image matching is an important research topic in the fields of computer vision and image processing. It has application in many fields such as military, industrial, medical, remote sensing, and so on. The development of sensor technology makes the image data obtained richer. Multi-sensor technology is applied to the image matching, which expands the application areas of the image matching technology.There are two main problems in multi-spectral image matching. One is the images obtained in different time have many changes such as rotation, scale, noise, distortion and so on. The other is the difference between images obtained by different sensor isn't ignorable. With the purpose of resolving those problems, this thesis gives a survey on the matching between visible images obtained in different time and the matching between the infrared and visible images.After analyzing the imaging models of infrared and visible images respectively, the characters and matching features between infrared and visible images are compared and summarized. A matching method based on the edge information was studied, which reduces the difference between the infrared and visible images through edge extracting. The technology of image pre-processing, edge extracting, noise and false edge detecting, similarity measures about infrared image are researched. By the stable performances in the matching of multi-spectral images, the method based on edges is chosen as our means.After analyzing and comparing the characters and the matching features of visible images obtained in different time, a matching method based on key points is studied. Many common methods of corner extracting are discussed and compared. An interest points extraction based on SIFT is described. Adding the distance rules to identify the error matching points make the error reduce. The experiments show that the method can keep invariance about rotation, scale, distortion, and noise.