摘要
针对模糊C均值聚类图像分割算法运算量大、难于实时处理的问题,提出了一种基于图形处理器的加速算法。通过分析模糊C均值聚类算法各阶段可以并行处理的运算部分,利用计算统一设备架构软硬件结构,分别将隶属度矩阵计算、聚类中心计算和像素按隶属度归类3个部分改造成适合图形处理器硬件并行运行的形式。实验结果表明,相对于CPU串行算法,基于图形处理器的加速算法效率提升明显。鉴于大多数图像处理算法均具有可并行处理的部分,利用图形处理器进行加速具有普适性。
In order to accelerate the segmentation algorithm of FCM（fuzzy c-means clustering）,an accelerating algorithm based on GPU（graphics processing unit） was proposed.Firstly,this method analyses the various phases of FCM algorithm which could be paralleled.Then,in order to adapt to the GPU＇s hardware architecture,this method transforms the computing of membership grade and clustering center and the classifying of every pixels according to the membership grade with CUDA（Compute Unified Device Architecture）.Experimental results show that the efficiency of the FCM segmentation algorithm accelerated by GPU is improved obviously compared with CPU＇s serial algorithm.In view of the parallel features of most image processing algorithms,the acceleration based on GPU is universal.
出处
《计算机科学》
CSCD
北大核心
2012年第1期285-286,294,共3页
Computer Science
关键词
模糊C均值聚类
图像分割
图形处理器
计算统一设备架构
Fuzzy C-means clustering
Image segmentation
Graphics processing unit
Compute unified device architecture
