直方图估计互信息在非刚性图像配准中的应用

The Application of Histogram-Based Mutual Information Estimation in Non-Rigid Image Registration

提出了一种对非刚性三维医学图像进行弹性配准的新方法，配准判据是基于二维联合直方图估计得到的两幅图像灰度之间的互信息，首先最大化全局互信息，实现两幅图像之间的全局仿射配准；然后将两幅图像的重叠部分均匀地划分成互为重叠的体积子块，再最大化对应体积子块的图像灰度之间的互信息，实现每对对应子块的局部刚体配准，将每个子块的中心作为一一对应的控制点。最后利用这些控制点对结合薄平板样条插值法实现全局弹性配准。

A new approach to the non-rigid registration of 3D multi--modality images is proposed with mutual information(MI) as its criterian. The mutual information is estimated with the aid of the 2D joint histogram. Based on the global affine registration by maximizing MI between the intensity values of the overlapping part of two images, the overlapping parts are divided into uniform sub--image blocks and the locally rigid registration is performed to produce a large amount of control point pairs by maximizing MI between the intensities of corresponding sub--images. The global elastic registration is achieved by combining those corresponding points with thin--plate spline.