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

摘要: 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) method, a fine method based on geometric constraints parallel architecture is put forward. First, order to provide for block registration, we use algorithm parallelly extract points calculate initial transform matrix. Then, segment overlapping original image pairs into several blocks. Finally, achieve blockwise matching eliminate through are implemented architecture. During affine invariance Mahalanobis distance screen duplicate matches mismatches. The experimental results show that this has eliminated generated same features. Moreover, accuracy speed been improved, proposed approach practical value.