The performance of a skull-based image registration is assessed for aligning intra-subject computed tomography (CT) images. Images with six levels of resolution are provided. The images with coarser resolution are resampled, followed by thresholding all images to extract the skull as the matching feature. Using the finest image as the reference, the skull-based registration is processed with sum of squared difference (SSD) criterion. In addition, images with partial contents are simulated and served as the floating image to investigate the interaction between different levels of resolution and partial contents. We conclude that a successful registration requires that more than half of the content should be preserved and the differences of image resolution between two images should be less than 16 times. © 2010 IEEE.
|出版狀態||已發佈 - 2010|