Comparative study of intensity based cost functions for automated satellite image registration

摘要

Choosing a suitable cost function for satellite image registration is always a challenging task. Satellite images poses a unique challenge in image registration field as images may consist of noise, clouds, highly undulating terrains etc. Registration of Satellite images lie in the category of multi temporal as well as sometimes multi modality cases. Till today there is no generalized solution suitable for all kind of cases. Choice of cost function plays an important role in image registration task. Different types of cost functions are available like intensity difference, correlation, entropy based methods etc. Here detailed exercise is done on existing similarity metrics, with a final conclusion that mutual information based similarity metric covers most of the cases with satisfactory results.