Performance Study of Optimization Methods for Intensity Based Automatic Satellite Image Registration

Abstract

Satellite Image registration plays an important role in remote sensing data processing and applied in wide variety of tasks such as image fusion, image overlay and change detection using different images of the same region. It is one of the challenging image processing tasks due to imaging by different sensors, at view angles or and at different times. Satellite images also poses unique challenges for registration with issues like cloud pixels, noise in the images, systematic errors, multispectral images, terrain induced distortions etc. It requires building an elaborate computational framework to handle specific problems of satellite image registration as the automatic image registration is very important requirement for voluminous data sets. Conventional approaches in satellite image registration involve a feature collection strategy manually or automatic, measuring similarity measures to find the best feature matches and further use spatial coordinates of the best matches to estimate a transform. Recent advances in medical image registration topic have suggested employing intensity based non rigid image registration framework that engages a sampling strategy, a similarity metric, a transform and an optimization procedure in an iterative manner. This procedure finds optimal transform parameters by maximizing the chosen similarity measure criteria and thus minimizing a cost function, providing a robust image registration framework. Satellite image registration can be treated as an optimization problem with the goal of finding the spatial mapping that will bring the two images into alignment. So a suitable choice of optimizer plays a key role in registration process. Ways of employing and comparing the performances of different optimization methods such as Evolution Strategy, Conjugate Gradient, Gradient Descent, Simultaneous Perturbation, Robbins-Monro, Adaptive Stochastic Gradient Descent, and Quasi Newton is reported here for the intensity based satellite image registration. Elastix, a public domain tool developed for doing intensity based medical image registration has been used in this study to perform non-rigid satellite image registration.