动态射线数字图像序列降噪算法及快速实现  

Denoising Algorithm for Dynamic Digital X-ray Image Sequences and Its Fast Implementation

作　　者：杨民[1] 孟凡勇[2] 梁丽红[3] 魏东波[1] 

机构地区：[1]北京航空航天大学机械工程及自动化学院,北京100191 [2]中国科学院过程工程研究所多相复杂系统国家重点实验室,北京100190 [3]中国特种设备检测研究院,北京100013

出　　处：《兵工学报》2011年第11期1331-1336,共6页Acta Armamentarii

基　　金：国家自然科学基金资助项目(60872080;61077011);中国科学院过程工程研究所多相复杂系统国家重点实验室开放基金资助项目(MPCS-2011-D-03)

摘　　要：为了实现动态射线数字成像(DR)系统的快速降噪,分析了X射线数字实时成像系统的噪声特点和原始的基于图像序列的NL-means降噪算法的不足,提出了一种改进的NL-means序列图像降噪算法。同时,为了解决NL-means降噪算法计算量大、运算速度慢的问题,利用可编程图形处理单元(GPU)并行计算和高速浮点计算特性,将图像映射为GPU中的纹理,采用多线程并行计算,使得NL-means算法在GPU中加速执行。实验结果表明,改进之后的算法有效地抑制了动态DR图像中的随机噪声,同时保留了图像的细节信息。GPU加速方法可以在不损失图像信息的前提下,实现实时降噪。In order to realize the fast noise reduction in the dynamic digital radiography（DR） imaging system,the system＇s noise characteristics and the disadvantages of the original NL-means denoising algorithm were studied.An improved NL-means noise reduction algorithm was proposed.Meanwhile,in order to solve its calculation complexity problem,graphic processing unit（GPU） was applied to make use of its high parallel and fast floating-point calculation abilities.In the implementation,the image was mapped to the GPU＇s texture and multi-thread parallel calculation was adopted.The experimental results showed that the improved NL-means noise reduction algorithm could effectively restrain the quantum noise and improve the resolution of DR image sequences.In addition,the acceleration with GPU could realize the real-time denoising without the loss of image information.

关 键 词：信息处理技术 射线数字成像 图像降噪 NL-means算法 图形处理单元 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]