Abstract: Aiming at the problem that common image feature matching algorithms will produce a large number of mismatched points in the image feature matching stage for images with parallax, a image feature matching algorithm combining accelerated-KAZE (AKAZE) algorithm with adaptive local affine matching was proposed. Firstly, the AKAZE algorithm was used to extract the feature points. Then, the binary descriptor M-LDB was used for description and the brute force matching was performed to generate coarse matching point pairs. Finally, the image-based affine transformation could provide the characteristic of strong geometric constraints, and adopted adaptive local affine matching to complete fine matching. The experimental results show that the algorithm has the effects of uniform extraction of feature points and accurate matching for image matching with rotation change, scale change, and viewing angle change at the same time. The number of correct feature points extracted is increased by average 1.66 times compared with SIFT algorithm, 1.08 times compared with SURF algorithm, 6.92 times compared with ORB algorithm and 1.23 times compared with GMS algorithm, respectively. It can meet the needs of image matching with large disparity.