摘要
结构相似度通常用来评估图像质量。当空间位置发生改变时,图像间的结构相似度也会随之发生变化,近年已用作单模态图像配准测度。对其进行适当修改,提出一种新的基于像素灰度的配准测度——修正的结构相似度函数(MSSIM),并且应用于不同分辨率MR/CT及MR/PET三维临床脑图像(由Vanderbilt大学提供)配准中,算法首先做质心对齐,然后利用Powell算法"由粗到精"对下采样图像配准,再由8点法评估配准质量。结果显示,此测度函数具有良好的配准性能,能够完全自动地达到亚像素级配准精度,鲁棒性较高,但运算速度也较慢。
Structural similarity is often used to assess image quality. The fuuction that structural similarity between images changes along with spatial location has been employed in one-dimensional image registration in recent years. We modified it and put forward a new registration metric based on voxel gray-modified structural similarity （MSSIM）, applying to three-dimensional brain image registration of MR/CT and MR/PET with different resolution （provided by the Vanderbilt University）. It starled from centroid alignment, and then sample image registration was perbrmed according to the Powe Algorithm, followed by quality assessment by eight-point algorithm. It turned out that the registration metric perlbrmed well and could reach the accuracy of sub-pixel registration fully and automatically. The robustness was high but the operation was rather slow
出处
《中国医学影像学杂志》
CSCD
北大核心
2013年第8期618-623,共6页
Chinese Journal of Medical Imaging
基金
鲁东大学横向基金项目(2010HX007)
关键词
脑图像
图像处理
计算机辅助
结构相似度函数
图像配准
Brain image
Image processing
computer-assisted
Structural similarity
Image registration
