Image stitching based on attention mechanism
Hongbin Wu, Yujing Gao, Ruoheng Ding, Lingchen Jin · 2022 IEEE 10th Joint International Information Technology and Artificial Intelligence Conference (ITAIC) · 2022
With the rapid development of science and technology, almost all the cameras of smart devices have the function of image stitching. However, the current more popular image stitching still has problems in accuracy. If we need to get more accurate image stitching, the requirements of the algorithm will be higher and more memory will be occupied in the matching process. To solve the above drawbacks, I plan to introduce an attention mechanism to extract feature points by selecting more important feature regions from the overlapping regions of the target object as subjective feature regions, thus reducing the memory occupation at runtime. The experiment uses a binocular camera as the acquisition tool for the target object. By using subjective feature regions to extract feature points to optimize the feature point extraction and matching process, after repeated experiments, compared with the traditional method of using subjective feature regions to extract and match feature points with the traditional method, not only the memory consumption is optimized, but also the accuracy of stitching can be improved by about 10%.