A Multi-Stage Astronomical Images Registration Based on Nonsubsampled Contourlet Transform

Abstract

In order to align the astronomical images with the characteristics of serious noise and smoothing edges, we propose an astronomical image registration based on the nonsubsampled contourlet transform (NSCT) and a new evaluation criterion to estimate the results of the registration. The registration algorithm includes coarse registration and fine registration. According to the shift-invariance of the NSCT, the approximate translations, which will be used to create the search windows of the fine registration, are obtained. Next, the local searches are operated in subband images, and then the feature points, which are extracted by using NSCT coefficients, are matched by utility of the gray correlation, and finally we can calculate the transformation parameters. The preliminary experimental results demonstrate the robustness and efficiency of the proposed algorithm in the noise suppression and the high registration accuracy which can achieve 0.2 pixels. Feature-based registration methods extract the edge features in the reference and sensed images, and then the parameters of the transform equation are obtained using these feature points, which are extracted by utilizing spatial relationships or similarity methods. The techniques of extracting the image edge features by using the edge detection operator or wavelet transform are proposed in the literature (4) (8), but the smooth edges cannot be effectively extracted. The algorithm based on the graph matching is proposed in the literature (9) and the random sample consensus (RANSAC) method is proposed in the literature (10), both of them are used to solve the optimal solution of the matching points. According to the characteristic of the astronomical images, whose change are slow and noise is relatively serious and feature structure is smoothing, a new astronomical images registration algorithm based on NSCT is proposed in this paper.