A Study Of SAR Image Registration Using Edge Information Combined With Gray-Level Information

Registration is a fundamental task in image processing used to (geometrically) match two or more images, for example, at different time, from different sensors or from different viewpoints. Registration is the foundation of many image processing and applications such as object recognition, change detection, 3-D modeling, and the effect of registration directly works on the following processing steps.There are two types in image distortion: those which are the source of misregistration and those which are not the source of misregistration, image registration is to find proper technique which is suitable for the former while eliminate the effect of the latter. There are a variety of registration method, but none of them are adapted to all images, so how to obtain a proper method to the image we are working on is the main task on image registration.Since synthetic aperture radar (SAR) imaging does not affected by weather conditions, so it is more and more widely used on many fields. But because of different imaging types, SAR image has large difference with optical image, techniques adapt to optical image registration can not be directly used to SAR image registration. So our aim is to find methods adapted to SAR image registration in this paper.First, we discuss on the key of feature-based registration methods: feature extraction and feature matching. On the one hand we analyze the problems when classic feature extraction methods used on SAR image; On the other hand, we design experiments to choose for a proper similarity criterion for SAR image, and use this criterion in the following SAR image registration method.Next, this paper describes a new feature-based method using edge information of typical objects combined with local gray-level statistical information to image-to-image registration. Generally, It is difficult to extract and matching feature in high precision using classic registration method due to speckle noise and the complex distribution of gray-level in SAR image. The method we designed first selects local area with typical object and extracts edge information from these object, then feature points is matched using a criterion combined with edge shape and gray-level information. In such steps, we obtain robust and steady control point using for image registration.Based on the method described above, experiments are taken on multitemporal Radarsat images. We obtain satisfactory results and validate the availability of this method on SAR image registration.