Incorporating spatial information into entropy estimates to improve multimodal image registration

We propose an approach to incorporate spatial information into the estimate of the entropy of an image by partitioning a high-dimensional space into regions over which a one-dimensional histogram can be indexed. This approach obviates the need for the construction of high-dimensional histograms and makes no assumptions on the form of the high-dimensional probability mass function. We use this partitioned estimate of entropy to compute image similarity measures, useful in image registration, and we examine the convergence properties of the registration process