Elastic Registration of Remote Sensing Images for Change Detection

Algorithms based on global rigid model can not resolve local geometric distortion problems caused by internal and external factors such as different remote sensing payloads, observation angles, times, and topography. Gobal algorithms restrict accuracy improvemenst in automatic registration and change detection in remote sensing images. In this paper, we present an elastic registration method based on a preliminary global Speed Up Robust Feature (SURF) affine registration method, local translation, and smoothing models. We constructed the weight function with normal density function of each pixel in the difference image to weaken errors of local translation paramters, caused by different radiation intensities of pixels f. Experiments using urban area data show that this method of image registration (including topography area) is more accuracate than a pixel, and is applicable and effective for change detection.