基于Fermi平台的双边滤波超声图像斑点噪声抑制并行处理算法

PARALLEL ALGORITHM WITH BILATERAL FILTERING FOR ULTRASOUND IMAGE SPECKLE NOISE SUPPRESSION PROCESSING BASED ON Fermi PLATFORM

对医学超声图像使用双边滤波技术不仅可以很好地抑制噪声,同时可以较好地保留组织结构信息。但由于该技术涉及大量的复杂计算往往难以满足实时处理的要求,为此依据Fermi架构技术特性设计完成一种斑点噪声抑制的并行处理算法。数据测试结果显示,采用Fermi架构的GPU处理在保证算法处理效果基础上获得了比较明显的加速效果。对于512×512的图像数据在迭代两次的情况下能够达到25 fps的帧率,速度提高了大约90倍。

Applying bilateral filtering technique in medical ultrasound imaging system can suppress noise very well; meanwhile the tissue structure information can also be well preserved. However, since this technique involves in massive complicated computation, it is usually difficult to meet the requirement of real-time processing. Therefore, in this paper we design and implement a parallel algorithm for speckle noise suppression processing according to the technical character of Fermi architecture. Numerical test results show that by using Fermi architecture, the GPU （graphics processing unit） processing achieves quite obvious speedup effect on the basis of maintaining processing result of the algorithm. It can achieve 25 fps frame rate for the image data with 512 ×512 size in circumstance of iterating twice, the speed increases about 90 times.