Image Mosaic Technology Research

The objective of image mosaic is to compose a panoramic image, from a set ofsmall-sized images with overlapped field of view about one common scene. Recently, it iswidely used for image processing fields and has become a hot research area.This paper systematically studies the current compared mainstream two kinds of imagemosaicing method. One is the image mosaicing method based on SIFT features. The other isthe image mosaicing method based on SURF features. Firstly, the basic process of imagemosaic is studied and the problems that exist in the splicing process are analyzed in detail. Inthe image preprocessing stage, we mainly solve the image geometric deformation and noiseproblem to make cell image sequence unscribbled. In the image registration stage, threedifferent feature extraction operators are introduced. They are SIFT operator, SURF operatorand MSER operator. We realize the first two algorithms with Visual C++2008and OpenCV.And the SIFT operator is optimized to make it have higher real-time feature. But we realizethe last one algorithm with matlab2008a and vlfeat library. And the MSER operator isanalyzed in the aspects of invariance. Then, according to the different feature extractionoperators, we use the SIFT operator and SURF operator on the panorama of seamless splicing.In this method the nearest neighbors of each feature are found by performing the prioritysearch with BBF algorithm, and the Random Sample Consensus (RANSAC) algorithm isutilized to decide the transformation parameters of the images stitching. According to thehomography matrix, the cell image sequence was projected in the uniform plane coordinatesystem. Last, we use multi-resolution fusion method to realize seamless integration.Experimental results show that SIFT algorithm and SURF algorithm can obtainsatisfactory effect in the panorama splicing aspects.