SIFT Based Image Matching Algorithm Research

Image matching is an important research topic in the fields of computer visionand image processing. This paper makes a deep research on image matching method,the main research content is: image pretreatment, image feature extraction andmatching of the basic theory and key technology.In order to improve the search precision, eliminate or reduce the impact broughtby the various error factors on matching performance, appropriate preprocessing on theimage has to be done before the image matching. This article studis the imagepreprocessing both in the image smoothing and the image enhancement, meanwhile, apretreatment method for image matching is presented,which is suitable for SIFT(ScaleInvariant Feature Transform)feature point extraction.There are kinds of method of image matching being describing in this paper, onMoravec, Harris, SUSAN, SIFT four kinds of classical operator for a comparativeanalysis of the experiment, it is concluded that SIFT algorithm has all these featuressuch as brightness, rotation, scale, affine transformation invariance, and the goodrobustness to noises.So this article chooses SIFT for operator feature extraction. But ithas many problems when the features are extracted and matched such as less matchingpoints, poor matching precision, low matching accuracy, easy to produce the repeatedmatching point, large calculated amount, worse real-time and so on.Finally, the above analysis of the SIFT algorithm is optimized based on theimproved, and KD-tree and quasi Euclidean distance based on the proposed imagematching method based on SIFT algorithm. Experimental results show that thealgorithm presented in this paper not only keeps the basic characteristics of SIFTalgorithm, but also has many advantages such as plenty matching points, highmatching accuracy, no duplicate points, matching efficiency higher merit and so on,provides precise matching point to generate subsequent product images mosaic andother related fields. At the same time, this method was applied to the layoutoptimization, achieved good results.