Image Registration using Geometric Deformable Model and Penalized Maximum Likelihood

W. Cho, Sunworl Kim, M. Lee, Soohyung Kim, S. Park, and C. Jeong (Korea)


Image segmentation, image registration, geometric deformable model, mutual information, penalized maximum likelihood, and ITK software.


In this paper, we propose new techniques using simultaneously two kinds of information which are deformable surface and voxel intensities. First, we segment the volume images by using the geometric deformable model and extract the feature points of shape from the boundary of segmented object. Second, we use the optimization measure of registration as the modified penalized maximum likelihood (MPML) function computed from two image intensities of the extracted feature points as well as the transformed feature points. The registration between two images is to find the transformation parameter providing the maximum value of the MPML computed from the two objects’ surface information. To evaluate the performance of the proposed registration method, we conduct various experiments and compare our method with existing ones such as MI and ML based registration. We first segment various 3D volume images and extract the boundary point as well as the cylinder surface of the segmented object. Then, we have evaluate robustness of our method under various degradation environments such as blurring, noise corruption and distortion, by comparing registration traces obtained from the original image and its degraded version. We can observe that the global maxima of a MPML function occur when the alignment of reference and floating images are perfectly completed.

Important Links:

Go Back