A Coarse-to-Fine Registration Method Based on Geometric Constraints and Parallel Architecture

Aiming at solving lengthy time-consuming registration and mismatch problems caused by images which have feature simliarity of local region in the traditional SIFT (Scale Invariant Feature Transform) registration method, a fine registration method based on geometric constraints and parallel architecture is put forward. First, in order to provide geometric constraints for block and fine registration, we use SIFT algorithm parallelly extract feature points to calculate initial transform matrix. Then, segment the overlapping region of original image pairs into several blocks. Finally, to achieve fine registration, blockwise SIFT matching and eliminate the mismatch through geometric constraints are implemented on parallel architecture. During the fine registration, we use geometric constraints and affine invariance of Mahalanobis distance to screen feature points and eliminate duplicate matches and mismatches. The experimental results show that this method has eliminated the mismatch generated by same local features. Moreover, registration accuracy and speed have been improved, the proposed approach has practical value.