Nonrigid Registration Combining Global and Local Statistics

Zhao Yi

UCLA Henry Samueli School of Engineering and Applied Science

RP-ID  :  070029

学科分类：计算机科学（综合）

美国|英语

来源: UCLA Computer Science Technical Reports Database

【 摘 要 】

In this paper we present a novel approach for the nonrigid registration of multimodal images, using normalized mutual information as a similarity criterion. The deformation is defined on the image lattice by a displacement field where no prior parameters need to be assumed. With a continuous representation of images and Parzen histogram estimators, we have developed the closed-form expressions of the criterion and its first-order variation with respect to the freeform deformation on the overlapping region. To further reduce the sensitivity to the changes in overlap statistics, spatial relationships are incorporated into the registration criterion through a weighted combination of global normalized mutual information and local matching statistics calculated from patch windows. Together with a viscous fluid regularizer, the deformation model allows for large spatial variations of overlap statistics. To characterize the performance of the algorithm, synthetic phantoms and clinical data are used in a validation study. The results suggest that the augmented normalized mutual information provides substantial improvements in terms of registration accuracy and robustness.