Feature Extraction And Matching Based On Improved KAZE Algorithm

Image feature extraction and matching is an important part in the field of image processing,As the best algorithm,high matching success rate and shorter algorithm running time is the pursue.The image matching process is divided into feature detection and extraction,feature descriptor and feature matching.For the first two steps of the algorithm are representative research,and the algorithms are compared,putting forward a optimization algorithm based on speeded-up robust features(Speeded-up Robust Features,SURF)and random sample consensus(Random Sample Consensus,RANSAC),use two-way matching to reduce the matching errors.After the zoom and rotation angles of the images,we can see that the matching success rate of optimized algorithm is higher than the original algorithm,and the matching speed is faster.In the traditional SIFT,SURF and other feature detection algorithm,use linear Gaussian Pyramid multi-scale decomposition,so that noise can be eliminated and extracted to a significant feature points.But Gaussian decomposition is sacrifice local precision,In the expense of Gaussian,not only not to retain the fuzzy object boundary information,in all scales of details and noise are smoothed.This will easily lead to fuzzy boundaries and details are lost.The nonlinear scale decomposition can solve this problem,the KAZE algorithm is a new multi-scale image matching method based on the nonlinear scale space,the algorithm uses nonlinear filter to construct nonlinear scale space,In fuzzy image processing,the detail and edge information is not affected.The algorithm of 2D images detection and matching based on the multi scale space.Compared with the SIFT and SURF algorithm,the computation complexity is reduced,and the robustness is improved.Finally,The process feature of the KAZE algorithm detection and the feature matching is improved,Change search method of original feature pointsand the original Euclidean distance matching method improved using the cosine angle,through simulation experiments,it can be concluded that the improved algorithm has better matching success rate,shorter algorithm running time.