Automated Mult-source Remote Sensing Image Registration Based On Combined SIFT And Gabor Features

Image registration is the process of overlaying images(two or more) of the same scene taken at different times, from different viewpoints, and/or by different sensors. [2] Remote sensing image registration is an indispensable part for remote sensing image fusion, change detection, image mosaicing and so on. For the increasing amount of remote sensing data, the cost of manual registration is prohibitive. So automated remote sensing image registration has become a research hotspot. multi-source remote sensing image registration is the most difficult problem in this field of research. This paper aims to improve SIFT algorithm to meet the requirements of remote sensing image registration.We classifies and outlines the current main-stream remote sensing image registration techniques, dig into SIFT-based remote sensing image registration techniques, and pointed out the flaws of applying it to multi-source remote sensing image registration are unrepeatable key points, unreliable dominate orientation assignment, SIFT descriptor and matching method.To make SIFT descriptor more reliable, we presents orientation restricted SIFT descriptor combined with Gabor features based on larger support region. Gabor features, which is widely used in image retrieval, reflects edge direction information and direction information of image texture. The uniqueness of SIFT descriptor is improved by combining with Gabor features.To make matching method more reliable, we presents a improved RANSAC(Random Sample Consensus) which exploits the reliability of matched pairs and the location of key points. By utilize those extra-information, our method can filter out outliers more effectively.