Area Based Approach to Registration of SAR Images with Sub-pixel Accuracy

Abstract

Synthetic Aperture Radar (SAR) images are often difficult to register with other images or maps of the same area due to speckle noise and salt and pepper texture. The Geo-reference of the image is conventionally realized by manual identification of Ground Control Points (GCPs) and image rectification. However as the tonal contrast present within SAR images changes in accordance to the dielectric properties which varies seasonally, making geo-referencing of temporal SAR images acquired for the same geographical area more difficult. In the present paper a robust technique based on area base matching algorithm is proposed to find out GCP correspondence. The area based algorithms works out a statistical similarity measure for a small window of points or template between two images. This technique provides good match when there is a tonal similarity between two images. The SAR images show drastic seasonal variations thus restricting use of this algorithm. However the edge images show a fair match and can be used for finding out a similarity measure it was observed that the edges remains fairly stable for a long time. This paper presents the procedure for geo-referencing temporal images acquired for the same region by automatic identification of GCPs on the edge image by template matching technique. The procedure to extract high quality edge contours from SAR images for template matching has also been presented.