A refined automatic co-registration method for high-resolution optical and sar images by maximizing mutual information

Abstract

The use of multisensors images for different applications, like change detection and image fusion, require an image to image registration. Actually, the image registration becomes a crucial task with the continuous increase in image resolution, especially in urban area. In this paper, a refined automatic Mutual Information based registration approach of spaceborne Synthetic Aperture radar (SAR) and optical image is proposed. The data set include TerraSar-X (1m) and Worldview-2 (0.5m) images and cover urban area of San Francisco. Our approach is divided into two steps. First, we use Discrete Canny edge detector to extract contours from the optical image, and Gabor-Wavelet filter to extract edge from the SAR image. Second, a rigid transformation (rotation and translation) is applied to the contours image obtained by Canny detector. Then, for each value of transformation, the Mutual Information is computed between this transformed image and the feature image outcome of application of Gabor-wavelet to the SAR image. The best transformation parameters are obtained when the Mutual Information is maximal. Also, in this work, we compared the proposed approach with the basic Mutual Information intensity based registration method. The results obtained demonstrated that our approach improve the registration accuracy comparatively to the basic Mutual Information intensity based registration method.