Robust Multimodality Image Registration Based on the Joint Intensity Scatter Plot

When it comes to multimodal image registration, mutual information (MI) seems to hold the biggest market share. However, MI has a number of serious flaws, not the least of which is the quantization effect caused by histogram binning. We propose a novel registration method that, like MI, is motivated by the distribution of points in the joint intensity scatter plot (JISP). However, we avoid the pitfalls of binning by basing our cost function on the actual point locations in the scatter plot. Regressor points and line segments migrate around the JISP, trying to minimize the sum of the squared distances of the scatter plot points from the collection of regressors. Placed within a multi-resolution framework, this elegant method is efficient and intuitive, and exhibits accuracy comparable to (or better than) mutual information.