A Fast Medical Image Registration Method Based on Nonlinear Correlation Measurement

In this paper a new approach to the problem of rigid body registration is proposed, using a concept of nonlinear correlation information entropy (NCIE) as the new matching criterion. The presented method applies NCIE to measure the correlation degree between the image intensities of corresponding voxel in both images. Registration is achieved by adjustment of the relative position until the NCIE between the images is maximized. However, unlike the mutual information, NCIE changes in the closed interval [0, 1], and around the extremum it varies sharply, which makes it possible that a threshold of NCIE can be used to boost the search to registration transformation. Using this feature of NCIE, we registered a rotated floating image with a reference image combined the variant step-size searching for further enhancing the searching speed. Our results demonstrate the proposed method can take care the differences in contrast between the floating and reference images. Meanwhile, as the mutual information, NCIE criterion can be achieved automatically and without any preprocessing