基于GPU的弹性图像配准方法  

GPU-Based Elastic Registration

作　　者：张加万[1] 杨甲东[1] 孙济洲[1] 张红颖[2] 

机构地区：[1]天津大学计算机科学与技术学院,天津300072 [2]中国民航大学航空自动化学院,天津300300

出　　处：《天津大学学报》2008年第8期946-950,共5页Journal of Tianjin University(Science and Technology)

基　　金：国家自然科学基金资助项目(60373061);天津市科技攻关培育项目(04310491R)

摘　　要：通常的弹性配准技术因其计算强度大,消耗时间长,难以满足实时应用的要求.新一代图形处理器(GPU)以其用户友好的可编程性和出色的并行计算能力,为解决该问题提供了新的途径.根据GPU的自身特点,以薄板样条插值作为变换模型,构建了弹性配准计算平台.对二维单模态和多模态的两组图像进行实验,结果表明,相比于CPU,利用GPU可以更为迅速地获得变换参数,对于大尺寸、高分辨率或者多局部形变的图像,GPU的处理速度超出CPU1个数量级以上.It is difficult to employ the elastic registration technique in real time application due to its computation cost. With user-friendly programmability and parallel computation,contemporary graphics processing units （GPU） provide a new strategy to solve the problem above. Based on the characteristics of GPU, a platform was proposed to deal with elastic registration using thin-plate spline as the transformation model. Experiments for 2D registration of monomodal and multimodal images were performed. Results show that by using GPU transformation can be obtained more rapidly than by CPU. Particularly, the processing speed of GPU is higher than that of CPU by over one order of magnitude when dealing with images of large size and high resolution.

关 键 词：弹性图像配准 图形处理器(GPU) 薄板样条(TPS) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]