The improved algorithm of remote sensing image registration based on SIFT and CONTOURLET transform

This dissertation put forwards an improved algorithm of image registration based on the combination of SIFT and CONTOURLET Transform. The first thing is that the image will be decomposed into low-frequency sub-band and high-frequency sub-band by using CONTOURLET Transform. The next step is to extract the image´s edge feature points by setting a reasonable threshold value to the high-frequency sub-band, and to realize the coarse matching of high-frequency sub-band according to the mutual correlation coefficient method. The last is to extract the SIFT feature points of the low-frequency and complete the coarse matching. In this way it can not only reduce the computation on a large scale SIFT, but also eliminate part of the noise in the original image. Therefore, it is more conducive to extract SIFT features. After the coarse matching is completed, it is necessary to choose the fine matching points to realize the image registration using the improved random sample consensus (RANSAC).