摘要
针对传统相似性测度易受灰度偏移场的影响而造成误配,以及单层P样条变换模型中通常无法准确选择初始化网格密度的问题,提出了多层P样条和稀疏编码的非刚性医学图像配准方法。该方法将稀疏编码作为相似性测度,首先把待配准的两幅图像划分图像块,然后使用K-SVD算法训练图像块得到分析字典并寻找稀疏系数,采用多层P样条自由变换模型来模拟非刚性几何形变,结合梯度下降法优化目标函数。实验结果表明,与单层P样条几何变换和sparse-induced、rank-induced相似性测度相比,所提方法能够准确地选择网格密度,并有效克服灰度偏移场对配准的影响,降低了均方根误差,提高了配准的精度和鲁棒性。
This paper proposed the non-rigid medical image registration method of multilayer P-spline and sparse coding according to two problems,the first one is the mismatch because the traditional similarity measure is easily affected by the gray bias field and the second problem is that single-layer P-spline transformation model is unable to accurately select the initialized grid density.This method regarded sparse coding as the similarity measure,at first,it divided two registered images into image blocks.And then used K-SVD algorithm to train them and thus obtain analysis dictionary.In addition,it found the sparse coefficients and used multilayer P-spline free transform model to simulate the non-rigid geometrical deformation.At last,it combined with the gradient descent method to optimize the objective function.Experimental results show that compared with the single-layer P-spline geometric transformation,sparse-induced and rank-induced similarity measure,the proposed method can accurately select the grid density and effectively overcome the influence of the gray bias field on the registration,reduces the root mean square error and improves the accuracy and robustness of registration.
作者
王丽芳
成茜
秦品乐
高媛
Wang Lifang;Cheng Xi;Qin Pinle;Gao Yuan(School of Data Science ＆ Technology,North University of China,Taiyuan 030051,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第8期2557-2560,共4页
Application Research of Computers
基金
山西省自然科学基金资助项目(2015011045)
关键词
图像配准
稀疏编码
多层P样条
梯度下降法
image registration
sparse coding
multilayer P-spline
gradient descent method
