Local Non-uniform Invariant Moments Described In Image Stitching Technology Research

The application of panorama image is becoming more and more popular as the development of computer technology and the continuing demanding of high quality life. There are two main ways to get panorama image, one is from a panorama camera, and the other is mosaicing from several common cameras. The former one can get high quality image at an expensive cost, and the later one is much less expensive but the image quality is not satisfying. So a compromise between the cost and the image quality is the purpose of image misaicing. There are two critical technologies in image misacting:image registration and image fusion. Improvements on both of them are the common aim of state-of-the-art study of image mosaicing.In this thesis, local feature point based image registration was discussed as its accuracy and computation was both suits to image mosaicing. There were three steps in image registration:feature detecting, describing, and matching, although, describing seemed more important than the others. So in this thesis, image describing was studied, and a new local non-uniform invariant moment descriptor—NUCFMs was proposed, and was proved to have good in image description especially for small image.And then, the NUCFMs based descriptor was applied in image registration. As it can differentiate feature points much easier, the registration accuracy and computation were both improved.Finally, in order to simplify the image fusion, improvement on IHS based fusion was proposed in this thesis:in the fusion course, I and S were included but H. Then an optimal edge was determined using greedy algorithm. Experimental results showed the combination of greedy algorithm and IHS based fusion can both have smooth color transition and invisibility edge between two stitching image, which made the final mosaiced image to be a seamless one.