A high-precision registration algorithm for heterologous image based on effective sub-graph extraction and feature points bidirectional matching

Abstract

Aiming at solving the problem of low matching accuracy caused by different imaging mechanism of heterologous image, we propose a novel image registration algorithm based on effective sub-image extraction and bidirectional matching for surf feature points. The algorithm adopts a coarse-to-fine matching strategy. Firstly, we transform the edge image into frequency domain through fast Fourier transform, and roughly estimate transform parameters using the cross power spectrum; secondly, we divide the images after rough matching into several sub-graphs, from which we will pick out the effective sub-graph in terms of normalized mutual information, then we match bidirectionally the feature points of effective sub-graph pair according to time domain features, thus obtaining accurate transformation parameters, completing the fine matching. Experimental results of heterologous images in different scenarios show that, the proposed algorithm effectively improves the registration accuracy which is up to sub pixel level.