海量数据流的提升小波变换并行算法研究  

Study for Huge Amounts of Data Streams Lifting Wavelet Transform Parallel Algorithms

作　　者：刘光敏[1] 陈庆奎[1,2] 王海峰[3] 

机构地区：[1]上海理工大学光电信息与计算机学院,上海200093 [2]上海理工大学上海现代光学系统重点实验室,上海200093 [3]临沂大学信息学院,山东临沂276000

出　　处：《小型微型计算机系统》2015年第2期343-348,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60970012)资助;教育部博士学科点专项科研博导基金项目(20113120110008)资助;上海教委创新基金重点项目(13ZZ112)资助;上海信息技术领域重点科技攻关计划基金项目(09511501000;09220502800)资助;山东省自然科学基金联合专项项目(ZR2013FL005)资助

摘　　要：提升小波变换算法在图像去噪中有广泛的应用,但是对于海量数据流该算法计算速度缓慢无法达到实时性.为了提高计算速度,提出一种基于图形处理器(GPU)的并行计算策略,把传统提升小波变换算法映射到CUDA编程模型,利用具有大规模并行计算特征的GPU作为计算设备,结合GPU存储器的优势实现了基于滑动窗口的提升小波变换并行算法.实验的测试结果表明,在现有的实验条件下,随着图像的增加,提升小波变换并行算法可以把计算速度提高50倍,效率提高明显.本文提出的方法也可以用其他图像处理算法的并行化.While lifting wavelet transform algorithm has wide application in image denoising, for the huge amounts of data the flow al- gorithm calculation speed isslow cannot achieve real-time performance. In order to improve the computing speed, this paper proposes a parallel computing strategy based on the Graphics Processor Unit （ GPU ）. To promote the traditional wavelet transform algorithm mapped to CUDA programming model,using GPU which has the characteristic of large-scale parallel computing as a computing de- vice, and then combined with the advantage of GPU memory implements the Lifting wavelet transform algorithm based on sliding win- dow. The results of experiment showed that under the present experimental conditions with the increase of the image, lifting wavelet transform parallel algorithm can increase the calculation speed 50 times,, thus the algorithm ＇s property would be significantly im- proved. The proposed approach can also use other image processing algorithms of parallelization..

关 键 词：提升小波变换 图像去噪 海量数据流 滑动窗口 CUDA 图形处理器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]