Research On Image Registration Algorithm Based On SURF And KAZE

Image registration technique is a research hotpot in image processing,which occupy a very important position.The algorithm is mainly divided into feature-based and region-based registration,the former is sensitive to gray transformation,causing the registration accuracy is not ideal,and huger calculation,lower rate;on the contrary,the latter has smaller calculation,better robustness,and a strong adaptability against to image offset and rotation.It has gradually become the main direction in the field of image registration,one of the most widely used registration algorithm is based on point feature.This paper has studied registration algorithms and shortcomings of classical algorithms,then extend the research aim at the different transformation for the performance of the algorithm.Through theoretical research,this paper proposes a corresponding improved algorithm to enhance the accuracy and precision of the algorithm.The main work of this paper is as follows:? According to the classical SURF algorithm and KAZE algorithm,the basic principle of the original algorithm,the steps of the algorithm and the registration performance are analyzed.? In order to solve the existing problems of traditional SURF algorithm,improve the accuracy and precision of the color image registration,a bidirectional adjacent to the matching color image registration algorithm is proposed.This algorithm improves the traditional SURF feature descriptor;the color information of the image is superimposed on traditional SURF feature descriptor which only contains gray information,constituting the improved SURF feature descriptor,to enhance the color information impact of matching,and can improve the matching accuracy.To improve searching efficiency and matching precision,a bidirectional FLANN matching algorithm is applied to image matching;then wrong matches are eliminated by improved RANSAC to further optimize the matching results.The experimental results show that the algorithm can improve color image registration accuracy and precision effectively,and has better adaptability and robustness of image transformation.? In order to improve the blurred image registration properties of image registration algorithm,an adaptive blurred image registration algorithm which integrating of nonlinear scale space and space cosine similarity is proposed.In this algorithm,the nonlinear scale space theory is applied to extract the local feature of the image,and the feature points of the image are extracted by the KAZE algorithm to construct the M-SURF feature descriptor.In order to seek the optimal transformation relations of reasonable quantity of matching points,.space cosine and adaptively threshold are used to match feature point.Finally,delete mistake matching points to improve the accuracy.The experimental results show that the algorithm can effectively improve the matching accuracy of blurred image registration,and has better registration accuracy and precision in comparison with the KAZE algorithm.