Multi-modal and Multi-temporal Image Registration in the Presence of Gross Outliers Using Feature Voxel-Weighted Normalized Mutual Information

A novel automatic registration algorithm based on new similarity measure called Feature Voxel-Weighted Normalized Mutual Information (FVW-NMI) is presented for accurate and robust multi-modal and multi-temporal image registration in the presence of gross outliers. Based on assumption of corresponding voxels having some common similar spatial information between multi-modal and multi-temporal images, feature measure maps for both floating and reference images are computed to estimate a normalized joint feature weight map of overlapped regions of the two images with registration transformation T iteratively changed. Each normalized weight value at every voxel of normalized joint feature weight map shows the spatial feature similarity between neighborhoods of corresponding voxels in overlapped regions of the two images and simultaneously indicates to what extent that these corresponding voxels potentially don't belong to gross outliers. Using both the normalized joint feature weight map and the original images, the FVW-NMI is iteratively computed to automatically exclude gross outliers in voxel based image registration. The results show that the method is sufficiently accurate and robust to gross outliers for multi-modal and multi-temporal image registration.