Abstract: Aiming at the problem that the traditional scale-invariant feature transform (SIFT) algorithm has a large amount of calculation in the process of image stitching, a fast SIFT stitching algorithm based on phase correlation and texture classification is proposed. Firstly, by using the phase correlation method, the overlapping regions of the input images to be stitched are roughly obtained. Secondly, images are classified with texture and the area with higher texture complexity is selected for SIFT detection. In order to improve the speed of texture classification, an interval classification approach is proposed. Finally, the feature points are matched only in the regions with the same texture complexity for different complexity texture regions. The experimental results show that compared with the traditional SIFT algorithm and the two existing improved SIFT algorithms, the improved algorithm in this work not only maintains good stitching quality, but also improves the average stitching speed by 68.46%, 20.45%, 41.83%, respectively. Therefore, the proposed algorithm has the potential application value in the field of high stitching efficiency requirements.

Key words: image processing, image stitching, scale-invariant feature transform, phase correlation; texture classification, interval classification