一种基于非下采样剪切波变换的医学图像配准方法

Medical image registration method based on non-subsampled shearlet transform

针对经典弹性配准方法在医学图像应用上计算复杂、方向信息不足的问题进行了研究,提出一种基于非下采样剪切波变换(NSSTR)的医学图像配准方法。利用剪切波能够对各个尺度、方向和位置实现较好定位的优势,先用该算法对医学图像作非下采样剪切波变换,得到各尺度、方向子带的剪切系数,然后对高频、低频的变换图像分别使用不同的配准方法,最后统一到变化网格上得到最终配准图像。实验结果表明,本方法与传统方法相比,不仅具有配准精度高、鲁棒性好的特点,而且计算效率更高。

The computational complexity of medical imaging and insufficient direction information are the drawbacks of classical elastic registration methods. Based on the study for the problem,this paper proposed a new medical image registration algorithm based on non-subsampled shearlet transform（ NSSTR）. In the method,it used the privilege that shearlet could locate each scale,direction and position. Firstly,the algorithm used non-subsampled shearlet transform on medical image to obtain scale,direction of subband shear coefficient,then adopted different registration methods for high frequency and low frequency images respectively. Finally,the algorithm got changing grid and registration image. Experiments show that the new proposed method outperforms traditional methods in the aspect of registration precision and robustness. Especially the algorithm is with higher computational efficiency.