Fermi平台下基于直方图匹配的超声图像斑点噪声抑制  

ULTRASOUND SPECKLE NOISE REDUCTION ALGORITHM BASED ON HISTOGRAM MATCHING WITH FERMI ARCHITECTURE

作　　者：何兴无[1,2] 周洪林[2] 蒋蕙竹[2] 雷静[2] 

机构地区：[1]成都师范学院网络与信息管理中心,四川成都611130 [2]成都农业科技职业学院电子信息分院,四川成都611130

出　　处：《计算机应用与软件》2012年第10期276-279,共4页Computer Applications and Software

摘　　要：在医学超声成像系统中斑点噪声是影响超声图像成像品质的重要因素。使用直方图信息可以较好地识别出斑点噪声和组织结构区域,从而进行自适应滤波抑制噪声。但这一处理涉及大量的复杂计算,使其成为临床实时成像系统中的一大性能瓶颈。为此研究并提出一种基于高性能计算平台Fermi架构GPU(graphics processing unit图形处理单元)的并行处理算法。数据测试结果显示,与基于CPU的实现相比,采用Fermi架构的GPU处理不仅可以得到完全一致和较好的图像去噪效果,而且可以获得良好的加速性能,基本满足实时系统的需求。对于512×512的图像数据能够达到228fps的高帧率,速度提高了大约115倍。In medical ultrasound imaging system, speckle noise reduction is the most important way to improve the ultrasound imaging qualities. With histogram information of the ultrasound images, the speckle noise can be distinguished from tissue structure region, which could be used to reduce the speckle noise by adaptive filter with preferable result. However, because of the massive computation involved in this filter technique, it has been the bottleneck for the clinical real-time imaging system in improving its performance. In light of this, we study and present a new parallel processing algorithm based on Fermi architecture GPU （graphics processing unit） for speckle noise reduction with histogram matching. Numerical test results show that compared with CPU-based implementation, the output of graphics processing unit （GPU） with Fermi architecture can not only obtain the image denoising effect fairly good and definitely the same as the one of CPU, but also achieves obvious speedup which meet the requirement of a real-time system, it achieves 225fps for the image size （512 x 512） which is 115 times faster than the CPU implementation.

关 键 词：高性能并行计算 斑点噪声抑制 直方图 CUDA 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]