IMAGE REGISTRATION ALOGORITHMS FOR SATELLITE AND REMOTE SENSING APPLICATION

Abstract

quotImage Registration is the process of overlaying images of the same scene taken at different newlinetimes, from different viewpoints, and/or by different sensors. It is a very important preprocessing newlinestep in the applications such as remote sensing, medical diagnosis, computer newlinevision etc. where the final information is derived based on the comparison of the images. newlineThe reason for the increased significance of image registration for satellite images is that newlinethe remote sensing is currently moving towards operational use in many important newlineapplications, for societal benefits as well as scientific study. The satellite images are multitemporal newline(taken at different dates), multisource (captured from multiple sensors), multispectral newline(captured at different frequency bands) or multimodal (obtained with different newlineacquisition modalities). Image registration for remote sensing is also difficult due to the newlinechallenges such as large image size, having nonlinear variations in intensity level, newlineatmospheric effects, noise, presence of clouds, occlusions etc. newlineBroadly there are two classes of approaches for image registration: area (or intensity) newlinebased methods and feature based methods. In the framework of area based methods, choice newlineof similarity measure and search strategy play significant role. Image registration using newlinemutual information as a similarity measure is investigated as it is best suited for newlinemultimodal images; but the computational complexity is challenging. An alternative newlineapproach to image registration is to transform the spatial information of image into another newlinetransform domain such Fourier transform and then to use the properties of the transform newlinedomain to estimate the registration parameters. Use of radon transform is investigated to newlineestimate the registration parameters, which founds to be close to the actual parameters and newlinerobust to noise as well. newlineIn feature based methods, basically there are four steps: feature detection, feature newlinematching, estimation of registration parameters and re-sampling. Due to its advantage