Improvement of SURF Feature Image Registration Algorithm Based on Cluster Analysis

The present study concentrating about image registration based on feature. With the shortcomings of low precision and limited applicability in the traditional registration algorithm by using random sample consensus method, an improved registration algorithm is proposed in this paper. Cluster analysis based on shared nearest neighbor is used to eliminate mismatching pairs. Initial matching pairs are used to construct two-dimensional vectors. These vectors are divided into different groups according to SNN similarity, and then the appropriate groups are selected as the final matching pairs to calculate parameters of the transform model. Experimental results show that this algorithm improves the accuracy, applicability and robustness of registration, and satisfies the need of image mosaic.