基于CUDA的快速中值滤波算法  

Fast Median Filtering Algorithm Based on CUDA

作　　者：吕亚飞[1] 贾堃阳[1] 

机构地区：[1]四川大学计算机学院

出　　处：《现代计算机》2011年第13期3-6,共4页Modern Computer

摘　　要：在众多的快速中值滤波算法中选取一种适合在CUDA平台上实现的算法。并针对GPU的运算特点．对算法进行很大的改进。改进后的算法采用纹理存储器存储数据源，共享存储器和寄存器存储中间运算结果．并通过同一block内的线程共享排序结果减少了排序过程中所需的比较次数．降低了算法的复杂度。实验结果表明改进后的快速中值滤波算法充分发挥了GPU强大的并行处理能力．对于分辨率为4096×4096的图像其运算速度是基于CPU实现的6597倍．可有效地应用在实时图像处理中。To realize median filtering algorithm efficiently on CUDA platform, uses a special fast median filtering algorithm. A big improvement has been made on the selected algorithm according to the features of GPU computing. The improved algorithm uses texture memory to store data sources, shared memory and registers to store operations result. And by the method that threads in the same block can share sorted results, the number needed to compare in the computing process is rapidly reduced. Experimental result shows that the improved fast median filtering algorithm makes full use of the parallel processing ability of GPU, with a speed which is at most 6597 times faster than the algorithm based on CPU, proves that it can be effectively applied in real-time image processing.

关 键 词：中值滤波 实时算法 GPU CUDA 

分 类 号：TN713[电子电信—电路与系统]