Researches On Image Mosaic Based The Constraints Of Distance

Image mosaic technology is an important research field of image processing. It isa technology that carries on the spatial matching to a series of image which areoverlapped partly with each other, and finally generates a seamless and big eyespotimage by merging these images. It has an important utility value in the fields ofremote sensing image processing, medical image analysis, synthesis of panoramicimages and so on.This thesis first introduces the significance of research, the present researchstatus, the applications domains and theory of image stitching technology. Then thisthesis discusses the basic steps of image stitching. It also deeply analyzes andsummarizes the key technologies of image stitching, include their advantages anddisadvantages. This thesis also proposes an improved algorithm based originalalgorithms. And then it compares the stability, noise resistance and time cost betweenthe improved algorithm and original algorithm by experiments.This thesis analyzes and researches major aspects of the image mosaic, dependingon the steps of image mosaic based corners. Firstly the Harris corners extractionprocess has been improved. According to the characteristics of feature extraction, inorder to enhance the speed of the operator extracting, this thesis proposes apre-extraction algorithm. And it removes the relatively low gradient pixels, in order tonot participate in the next step of the extraction, so the operator can effectivelyimprove the extraction rate. Then to aiming at the problem of the uneven distributionfor the extracted corners, the algorithm all block the image, and then select cornersusing the extraction function based the image contrast proposed in this paper, tochoose the high quality corners,and make the extracted corner reasonable distribution.It is conducive to the follow-up image registration process. After the corner extractionis completed, this thesis analyzes the characteristics of image registration. A cornermatching method of the constraints of distance combining local and globalinformation is proposed. This method combines the image characteristics of local andglobal, with good quasi effect. The experiment results show that the methods canachieve accurate matching of no unmatched corners. After image registration, theRANSAC algorithm estimates the optimal transformation matrix. At last, this thesisadjusts two fusion images contrast based the image contrast, and fuses them using thefusion method of the effective weight based Euclidean distance. Thereby it caneliminate the image overlap seams, to achieve the requirement of seamless mosaic.In conclusion, this research work involves all major aspects of the image mosaicprocess, and the experimental results of each researched algorithm are intending andgood.