Support vector machine density estimator as a generalized parzen windows estimator for mutual information based image registration

Abstract

Mutual Information is perhaps the most widely used multimodality image registration method. A crucial step in mutual information is the estimation of the probability density function (pdf). In most cases, the Parzen window estimator is employed for this purpose which results in an excessive computational cost. In this paper we demonstrate that replacing the Parzen density estimator with a Support Vector Machine (SVM) density estimation will result in a significant reduction of the computational time. We verified this by registering 2D portal images to DRRs (digitally reconstructed radiographs) projected from 3D CT volumetric data.