基于SURF和光流场的医学图像配准技术研究  

Research of Medical Images Registration Based on SURF and Optical Flow

作　　者：苏孟超 李克伟 张聪炫[1,2] SU Meng-chao;LI Ke-wei;ZHANG Cong-xuan(School of Measuring and Optical Engineering,Nanchang Hangkong University,Nanchang 330063,China;Key Laboratory of Nondestructive Testing(Ministry of Education),Nanchang Hangkong University,Nanchang 330063,China)

机构地区：[1]南昌航空大学测试与光电工程学院,南昌330063 [2]无损检测技术教育部重点实验室(南昌航空大学),南昌330063

出　　处：《南昌航空大学学报（自然科学版）》2018年第4期16-23,共8页Journal of Nanchang Hangkong University(Natural Sciences)

基　　金：国家自然科学基金(61772255;61462062);江西省创新驱动"5511"工程优势科技创新团队(20165BCB19007);江西省优势科技创新团队计划项目(20152BCB24004);航空科学基金(2016ZC56005);江西省青年科学基金(20171BAB212012);江西省重点研发计划项目(20161BBE50080);江西省教育厅青年科学基金(GJJ150706)

摘　　要：提出了基于SURF算法和光流场模型相结合的多模态医学图像配准算法。由于传统光流场模型具有灰度一致性的约束条件以及多模态医学图像的结构差异较大,导致算法对于多模态医学图像无法取得较好的配准结果。首先引入直方图规定化算法作为医学图像预处理,解决多模态医学图像中灰度差异较大的问题;然后根据多模态医学图像结构差异较大的特点,本文采用分级配准的思路,将SURF算法和传统光流场算法结合起来可以获得较好的配准结果;最后采用CUDA并行加速的方法对本文算法进行GPU加速。研究结果表明:本文算法具有较高的准确性和更好的鲁棒性。An improved registration method of multimodal images combined optical flow and Speeded-Up Robust Features (SURF) is presented in this paper. As a noteworthy limitation,the traditional registration methods based on optical flow are not suitable for the multimodal images with large differences in structure. Firstly,histogram specification is used to transform modal of images before registration,the similar modal of images are beneficial to the following image registration. Then,in order to solve the problem of structural differences in image registration,an improved registration method of multimodal images combined optical flow and SURF is presented in this paper. Finally,the high performance computing technology based CUDA was proposed in this paper. Experimental results show that the measure indexes of our method are more competitive compared with the other methods,which indicate the proposed approach has the higher accuracy and better robustness.

关 键 词：多模态 图像配准 直方图规定化 SURF 光流场 CUDA并行计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]