Modified ICP Image Registration Algorithm Based on Spatial Structure Constraint

Abstract:

A modified iterative closest point(ICP) image registration algorithm based on spatial structure constraint is proposed to overcome the matching ambiguity of remote sensing image registration caused by outliers. This algorithm combines the similarities of spatial structure and feature to determine matching matrix of feature points, among which the similarity of feature is achieved by a local feature descriptor while the similarity of spatial structure is calculated by spatial coordinates of feature points. Different to current structure based algorithms, the similarity of spatial structure contains not only spatial distance of corresponding feature points but distance of neighboring ones. And in matching process, the matching pairs are determined by a bidirectional matching criterion from the view of reference and sensed images. Experiments on real remote sensing images of different characteristics show that this algorithm can enhance registration accuracy.

Key words:image registration; iterative closest point(ICP) algorithm; spatial structure constraint; similarity matrix