基于隐含形状表示和边缘信息融合的非刚体图像配准  

Nonrigid Image Registration Based on Implicit Shape Representation and Edge Information Fusion

作　　者：廖秀秀[1] 于慧敏[1] 扬威[2] 

机构地区：[1]浙江大学信息与电子工程学系,杭州310027 [2]浙江天正信息科技有限公司(浙江省计算机研究所),杭州310006

出　　处：《中国生物医学工程学报》2008年第3期340-346,共7页Chinese Journal of Biomedical Engineering

基　　金：浙江省重点科技计划项目(2006C21035)

摘　　要：本研究提出基于隐含形状表示和边缘信息融合的多分辨率网格非刚体图像配准算法,使用从全局到局部的层次变换模型覆盖整个变换域,解决有较大局部形变的图像配准问题。首先用隐含形状表示图像的外部轮廓,将轮廓作为距离函数的零水平集隐含地嵌入到高一维的距离变换空间,在该隐含嵌入空间中使用互信息方法,实现了一个具有平移、旋转、尺度不变性的全局配准框架,对齐图像外部轮廓。然后选择基于B样条的多分辨率网格FFD模型进行局部配准,兼顾了结果精确度和计算效率。算法采用了与图像边缘信息融合的方法,强调图像边缘信息在配准中的贡献,得到平滑、连续且保证一对一映射的变换域。最后将该算法分别应用于脑部MR、CT图像的配准,得到令人满意的效果。A nonrigid image registration algorithm based on implicit shape representation and multi-resolution grid with edge information fusion was presented in this paper. A hierarchical model including both global and local transformations was used to perform the registration for the dense local deformation. At first, the Euclidean distance transform was used to embed the extern contour of the image as the zero level set of a distance function in a one dimension higher distance transformation space, and in the implicitly embedded space, the mutual information method was used to implement a global registration framework that was invariant to translation, rotation and scaling to align the extern contours of the images. Then a B-spline-based free-form deformation （FFD） model with multi-resolution grid was chosen for local registration. The multi-resolution method took account of the precision of the results and the computational efficiency. The fusion of edge information was used in the algorithm to emphasize the distribution of edge information in image registration. The recovered deformation field was smooth, continuous and guarantees a one- to-one mapping. Satisfied results have been acquired while using this algorithm in the registration of brain MR and CT images respectively.

关 键 词：非刚体图像配准 隐含形状表示 互信息 多分辨率网格 信息融合 

分 类 号：R318[医药卫生—生物医学工程]