Research On Image Feature Extraction And Matching Based On KAZE

Computer vision and image processing are the present study focus,in which image feature extraction and matching technology is the precondition of image analysis and recognition.Image feature extraction simplifies high dimensional image data and extracts key information(keypoint)of image by computer language.It is the basis and premise of image matching.With the continuous development of image processing technology,image feature extraction and matching has already become the foundation of pattern recognition,artificial intelligence,data mining,etc.KAZE algorithm is a kind of image feature detection algorithm based on nonlinear diffusion filtering,which can reduce boundaries and detail loss,obtaining superior localization accuracy and distinctiveness while preserving scale invariant property.Based on KAZE algorithm,this thesis proposes an improved image extraction approach by adopting PCA(Principal Component Analysis)and choosing suitable proportion of principle component to eliminate noise and redundant descriptors.Experiment results show that the amount and information of descriptors are better than KAZE.This thesis also adopts FLANN(Fast Library for Approximate Nearest Neighbors)and RANSAC(Random Sample Consensus)methods to match descriptors and eliminate wrong and repeated matches.Experiment results show that this approach increases the amount of effective descriptors,improving the actual match performance and the match rate is also better than KAZE as well as other traditional methods while retaining the original distinctiveness and accuracy.