采用PCA和PSNR的医学图像配准  

Medical image registration based on PCA and PSNR

作　　者：潘梅森[1,2] 汤井田[1] 杨晓利[1] 

机构地区：[1]中南大学信息物理工程学院生物医学工程研究所,湖南长沙410083 [2]湖南文理学院计算机科学与技术学院,湖南常德415000

出　　处：《红外与激光工程》2011年第2期355-364,共10页Infrared and Laser Engineering

基　　金：湖南省"十一五"重点建设学科--光学基金资助项目;湖南省自然科学基金资助项目(07JJ6141);2009湖南省教育厅优秀青年基金资助项目(09B071)

摘　　要：医学图像配准是医学图像融合的前提,是目前医学图像处理中的热点,具有重要的临床诊断和治疗价值。利用图像矩计算医学图像的质心,根据图像坐标矩阵主分量分析(PCA)的特征矢量旋转不变性,计算参考图像和浮动图像的倾斜角,获得配准初始值;在寻找最优几何变换参数时,采用峰值信噪比(PSNR)作为参考图像和浮动图像的相似性测度。实验时,采用单纯形法作为多参数寻优方法。实验结果表明,该方法计算简单,运算量少,配准速度快,精度高,而且克服了容易陷入局部最优的问题。该方法既可用于单模态图像配准,也可以用于多模态配准。Medical image registration, as the prerequisite for medical image fusion, is a hot field in the medical image processing and commonly used for clinical diagnosis and treatment. By means of computing the medical image moment, the centroid was obtained. According to the rotational invariance of the eigenvectors of principal component analysis （PCA） of the matrix of the medical image coordinates, the rotation angles of reference and float images were computed respectively, the initial values for registering the images were obtained. When searching the optimal geometric transform parameters, peak signal-noise ratio （PSNR） was selected as the similarity measure between the reference and floating images. In the experiments, the simplex method was chosen as multi-parameter optimization one. The results show that the proposed method has fairly simple implementation, low computational load, fast registration and perfect registration accuracy. In addition, it can effectively avoid trapping in the local optimum. It is adapted for both mono-modality and multi-modality image registrations.

关 键 词：图像配准 主分量分析 质心 医学图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]