Feature Matching Algorithm Research And Improvement Based On Local Image Content

The core part of the machine target recognition technology is dependent on the imagematching technology. Meanwhile, image matching technology is also the focus on the imageprocessing in current research direction.Firstly, In this thesis, image matching technology is as the breakthrough point, bystudying the basic principle knowledge of image matching technology and based on thefeature information of image matching algorithm, an overcome the environment or thebackground changes, image scale transform, geometric transform and the outside or insidenoise efficiently, fast high-quality algorithm can be found out. The application in static objecttarget recognition, the analysis of the common image feature point matching algorithmperformance contrast and preferred choice based on the SIFT（Scale Invariant FeatureTransform） algorithm is put forward basis algorithm in this thesis.Secondly, Improvement strategy mainly have two directions to improve the SIFTalgorithm. First, by analyzing the problems existing in the SIFT feature points detectionalgorithm, such as: too much useless feature points are extracted; generates a high dimensionfeature vector. In feature point detection phase, these two problems is enough to causealgorithm obtain high cost of computational, high degree of complexity, and reduce the laststage of matching algorithm efficiency. Second, analysis problem that traditional Euclideandistance matching method by classifying SIFT, although the matching point matchingcomplexity is not high but prone to error, reduce the matching success rate affect the efficiencyof matching algorithm.Finally, As above reasons, the SIFT algorithm can be improved in three aspects. First,introduce screen pixel, reject useless feature points area, reduce the burden of detectionprocess of feature points. Second, introduced principle of concentric circles make the keyfeatures of vector dimension reduction that cause reduce late phase high matching processcalculation, improve the utilization rate of feature points, the name of improved scheme isSP-SIFT feature point detection algorithm. Third, introduced EANSAC（Euclid RandomSample Consensus）algorithm principle of the quasi linear confirm reasonable matching area,reject errors caused by Euclidean distance to match point, once again, comprehensive improve the precision and efficiency of matching algorithm. Finally, the improved algorithm is appliedto a static image feature recognition, which can achieve a good effective.