ABSTRACT
This chapter focuses on a phenomenon known as interpolation induced artifacts which has been reported to limit the performance of mutual information based registration techniques in many practical applications like three-dimensional (3D) brain image registration in medical imaging and multitemporal image registration in remote sensing. It shows the typical artifact patterns encountered in the mutual information (MI) registration function in one dimension: through linear interpolation, and through partial volume interpolation (PVI). The chapter introduces the rationale of the maximization of the mutual information criterion and reviews two widely used joint histogram estimation methods for MI calculation, namely linear interpolation and PVI, as well as a simple optimization method to actually maximize the MI measure. One major limitation of implementations of the MI criterion through either linear interpolation or PVI is the occurrence of periodical extrema in the registration function under certain conditions.
