An Improved PCA-SIFT Algorithm for Image Mosaics

Aiming at the problem that the SIFT algorithm does not fully consider the distribution of feature points in the image splicing and the calculation is complex and takes a long time, an improved PCA-SIFT algorithm is proposed. The algorithm firstly introduces an improved non-maximum suppression method to optimize the initial feature points so as to obtain a more even distribution of feature point sets. Then the 64-dimension SIFT descriptor is extracted based on the circular neighborhood, and the descriptor is further reduced using PCA to reduce the data complexity of the descriptor. Finally, the BBF search strategy based on K-D tree was introduced. The RANSAC was used to eliminate the false matching points, which improved the matching speed and matching accuracy. In the 10 sets of image stitching experiments, the stitching speed of this algorithm is 1.6~2.2 times that of the traditional SIFT algorithm. Experimental results show that the proposed algorithm has higher accuracy, better robustness, and stronger real-time performance.