Abstract:

Aiming at solving the insufficiencies of the SIFT registration algorithm such as large calculation, error matching when generating feature vectors and doing feature matching, we propose an optimized SIFT registration algorithm. The optimized registration algorithm firstly sharpens the edge of images by bringing in the Laplacian operator and extracts the characteristics of block images via the projection entropy of image unit information. Then feature points are matched according to the minimum Euclidean distance of entropy vector projection. Finally, the improved random sample consensus algorithm is adopted to eliminate error matching. We apply the optimized algorithm in panoramic mosaic and experimental results show that, the optimized registration algorithm exceeds the SIFT registration algorithm, which not only effectively improves the efficiency, but also reduces error matching, achieving good matching effect.

Key words: panoramic mosaic;image registration;SIFT algorithm;projection entropy of unit information