OPTICAL-TO-SAR IMAGE REGISTRATION BASED ON GAUSSIAN MIXTURE MODEL

Abstract

Image registration is a fundamental in remote sensing applications such as inter-calibration and image fusion. Compared to other multi sensor image registration problems such as optical-to-IR, the registration for SAR and optical images has its specials. Firstly, the radiometric and geometric characteristics are different between SAR and optical images. Secondly, the feature extraction methods are heavily suffered with the speckle in SAR images. Thirdly, the structural information is more useful than the point features such as corners. In this work, we proposed a novel Gaussian Mixture Model (GMM) based Optical-to-SAR image registration algorithm. The feature of line support region (LSR) is used to describe the structural information and the orientation attributes are added into the GMM to avoid Expectation Maximization (EM) algorithm falling into local extremum in feature sets matching phase. Through the experiments it proves that our algorithm is very robust for optical-to- SAR image registration problem.