Research Of Image Registration Based Oncross-correlation And Point Feature

Image registration is the process of matching or overlay images which are acquired at different times, different conditions(camera angle, exposure intensity, etc.) or different acquisition equipment. In order to solve the problems in two different application areas, this paper improves registration algorithms NCC, SIFT and proposes PNCC, FSIFT algorithm.1) In view of the high accuracy and speed of vision Measurement of circular saw blade, the PNCC algorithm is proposed. NCC does not need to extract features and suits for solving template matching problem of vision Measurement. In order to overcome shortcomings of large amount of computation. This paper combines the PSO with the NCC. However, PSO is easy to converge at local optimum. In order to avoid this defect, first of all, we introduce into attached particle swarm based on region in order to guide particle swarm converge to global optimal solution quickly. Secondly, Adding in blacklisting mechanism so as PSO has ability of jumping out local optimum. Finally, increasing random disturbance operators for the purpose of expanding scope of the search area of local optimum. Experimental results show that PNCC has shorter runtime with higher accuracy. And achieves high accuracy in vision measurement of circular saw tooth width and pitch.2) In view of the high precision and real-time requirements of aerial image mosaic, the FSIFT algorithm is proposed. SIFT need to extract features and suits for mosaic of aerial image. But the keypoints and local feature vector extracted by SIFT has weak robustness. So we use FAST in order to enhanced the unique of feature vectors. Geometric constraints of orientation difference, scales difference and orientation of matching connection line are used to remove incorrect matches. The experimental results show that the modified SIFT reduces mismatched rate and running time effectively. And achieves good result in aerial image mosaic.