Improved Nonsubsampled Contourlet Transform for Multi-sensor Image Registration

Homologous feature point extraction is a key problem in the multi-sensor image registration. In this paper, a new feature point extraction method using nonsubsampled contourlet transform (NSCT) and an adaptive shrink operator (ASO_NSCT) for multi-sensor image registration is proposed. Moreover, this proposed feature matching is different from the traditional feature matching strategies and is performed using a similarity measure computed from neighborhood circles in low-frequency bands. Then, a number of reliable matched couples with even distributions are obtained, which assures the accuracy of the registration. Applications of the proposed algorithm to different optical images as well as optical and synthetic aperture radar images show that, in each case, a large number of accurate matched couples could be identified. Additionally, the RMSE patterns are analyzed and comparisons of the parameters are carried out between the registration models and the actual ground structures, which further demonstrates the effectiveness of the proposed algorithm.