Image Mosaic algorithm based on improved AKAZE
Lili Qian, Hua Li, Mingna Xu · 2022 3rd International Conference on Computer Vision, Image and Deep Learning & International Conference on Computer Engineering and Applications (CVIDL & ICCEA) · 2022
In the process of image Mosaic, traditional image matching algorithms have high time complexity, too many low-quality redundant feature points extracted, and feature descriptors are not accurate enough to describe key points, leading to low matching accuracy and low Mosaic efficiency. An image registration algorithm based on improved AKAZE was proposed. First, phase correlation method is used to estimate the global coarse displacement motion parameters of matched images and calculate the coincidence range between images. Then, AKAZE algorithm is used for feature detection, and Opponent-LATCH color feature descriptor with color information is introduced to describe the feature points. Finally, KNN and Bidirectional matching principle is used for rough matching, and PROSAC algorithm is used for fine matching. Experimental results show that the proposed algorithm has better robustness and better matching accuracy and running time than traditional algorithms.