Improved SIFT Algorithm For Image Matching Research

With the continuous development of computer technology, image matching plays a keyrole in the field of computer vision and pattern recognition, which has also a lot of basicproblems in image analysis and processing tasks, so the image matching technology is ofgreat significance and practical value. At the moment, the image matching technology iswidely applied in fingerprint recognition, image mosaic, medical diagnosis,industrialproduct testing and other fields.Although at present there are many kinds of feasible image matching methods, but ingeneral direction fthe image matching method can generally be divided into two categories:region-based image matching method and feature-based image matchingmethod,region-based matching methods can also be subdivided into matching methodwhich based on gray scale or transform domain. This paper mainly researched thefature-based image matching method of three typical image matching algorithm: SIFTalgorithm, PCA-SIFT algorithm and SURF algorithm; First of all, it is carried on researchand analysis detailly that these three kinds of typical algorithm principle and algorithmsteps, then the comparison of SIFT or SURF experimental algorithm with translation,rotation, scaling, and different illumination transform image matching and it is analyedrespectively from the matching accuracy and efficiency of algorithm. At the same time, inview of the SIFT algorithm, the computation is complexity and image matching has the lowefficiency and other defects. This paper puts forward an improved matching based on theprocess of search strategy to improve the matching efficiency of SIFT algorithm.Search strategy in the process of image matching plays a key role, with the traditionalSIFT algorithm in the process of image matching, compared to classic KD tree searchstrategy,it is proposed establishment of multiple different random KD tree structure andconduct search simultaneously in this paper. It shows up a good matching performance, theexperimental results show that compared with the standard of the SIFT algorithm isproposed in this paper the improved image matching based on SIFT algorithm not onlyimproves the efficiency of image matching, but also maintaining a higher accuracy ofimage matching.