Automatic registration of multi-temporal remote sensing images based on nature-inspired techniques

Abstract

In this article, we present nature-inspired techniques for automatic image registration of multi-temporal satellite images. Multi-temporal satellite image registration is becoming increasingly important to aid in flood damage assessment. We consider two images in the registration process: one before-flood image and another during-flood image. The objective is to maximise the similarity metric (of these two images) using information theoretic measures such as mutual information (MI). The maximum MI would imply that the images are better registered. The function of these metrics for transformation parameters is generally non-convex and irregular and, therefore, makes it difficult to use standard optimisation methods for the global solution. In this study, nature-inspired techniques – genetic algorithm (GA), particle swarm optimisation (PSO) and firefly algorithm (FA) are used to search for the maximum MI. The multi-temporal images – Linear Imaging Self-Scanning Sensor III (LISS III) image (before flood) and Synthetic Aperture Radar (SAR) image (during flood) and Moderate Resolution Imaging Spectroradiometer (MODIS) images (before and during flood) are used to demonstrate the performance of the proposed image registration approach. From the results obtained, we compare performance evaluations and conclude that nature-inspired techniques are accurate and reliable in solving satellite image registration.