Study On SAR Image Registration Based On Regions And Improved Comprehensive Learning Particle Swarm Optimization

Synthetic Aperture Radar(SAR) is an active microwave sensor, and has the characters of all-weather, all-time, penetrating the surface and clouds, high resolution and side looking, and so on, thus it has been more widely used in remote sensing missions because the imaging conditions of SAR are not affected by weather or visibility. There are some geometric deformation between the images obtained in the same area at different times and at different perspectives for the same SAR system. It is a prerequisite to solve the problem for other subsequent SAR image processing including fusion, change detection, feature classification. Therefore, SAR image registration is a very important research topic in the field of SAR image processing.This thesis is concerned on the bi-temporal SAR images registration by region matching and optimization algorithm. The contributes of this thesis are two fold:1. The bi-temporal images are decomposed by wavelet transform at first. Then, the sensed image and its wavelet approximation component are divided to some saliency region based on spectrum residua approach. And the saliency regions are fitted into ellipses. Finally, the region matching is performed by rough searching and fine searching the corresponding regions in the reference image at every layer by the comprehensive learning particle swarm optimization successively. Thus, the registration in every layer is a coarse-to-fine process, which effectively improves the efficiency, accuracy and stability. The final transformation parameters are calculated by the endpoint coordinate of long axis and short axis and the center of the matching elliptical areas. This proposed method has been validated by testing on three couple of bi-temporal images of Yellow River Estuary taken by Radarsat-2 in 2009 and in 2008, respectively. Experimental results demonstrate the effectiveness of this method.2. The sensed image and the reference image are divided to some saliency regions based on spectrum residua approach respectively, at first. Then, the region matching is performed well by measuring the similarity of the couple of regions, which the similarity is defined by sorting the mode of gray histogram of the regions. During the searching by our improved, the ratio image entropy is proposed for measuring the similarity of the sub-images( i.e. the saliency regions), and finally, with the help of different coordinates of the feature points in sub-images and the original images, the registration parameters of the sub-images are converted into that of the original images. The proposed method has been validated by testing on four bi-temporal images of Yellow River Estuary taken by Radarsat-2 in 2009 and in 2008, respectively. Experimental results demonstrate the effectiveness of this method.