自然科学版 英文版
自然科学版 英文版
自然科学版 英文版
自然科学版 英文版
英文版编委
自然科学版 英文版
英文版首届青年编委

您目前所在的位置:首页 - 期刊简介 - 详细页面

中南大学学报(自然科学版)

Journal of Central South University

第48卷    第10期    总第278期    2017年10月

[PDF全文下载]    [Flash在线阅读]

    

文章编号:1672-7207(2017)10-2691-06
基于GPU的小尺寸FFT在实时图像复原中的优化
严发宝1,苏艳蕊1,赵占锋2,左颢睿3,柳建新4

(1. 山东大学(威海) 机电与信息工程学院,山东 威海,264209; 2. 哈尔滨工业大学(威海) 信息工程研究所,山东 威海,264209; 3. 中国科学院 光电技术研究所,四川 成都,610209; 4. 中南大学 地球科学与信息物理学院,湖南 长沙,410083)

摘 要: 为满足跟踪识别系统对图像复原的实时性需求,在图形处理器(GPU)上进行高效实现小尺寸二维FFT的优化策略研究。首先对二维FFT算法进行分析,根据图形处理器的特点,提出基于图形处理器的并行执行模型。基于该模型,从算法的复杂度、跳转指令的数量、共享存储器的访问冲突以及共享存储器的访问延迟及图形处理器的利用效率这4个方面进行优化策略的研究,提出相应的优化方法。在图像复原的实验中,先对基于GPU的小尺寸FFT优化方法与基于CPU的MATLAB传统算法进行计算精度对比,然后基于4种不同尺寸的图像在相同的GPU平台上再与NVIDIA公司提供CUFFT函数库复原算法进行计算效率对比。研究结果表明:该优化方法提供的图像复原算法复原效果好,与MATLAB效果图比较人眼观察不出差异;在计算速率上,提出的优化方法能够在19.6 ms内复原1帧128×128灰度模糊图像,计算速度与直接采用CUFFT函数库算法相比提高约1.8倍。

 

关键词: 图形处理器;小尺寸FFT;图像复原;并行优化;实时处理

Optimization on FFT of small size in real-time image restoration based on GPU
YAN Fabao1, SU Yarui1, ZHAO Zhanfeng2, ZUO Haorui3, LIU Jianxin4

1. School of Mechanical, Electrical & Information Engineering, Shandong University, Weihai, Weihai 264209, China; 2. Institute of Information Engineering, Harbin Institute of Technology, Weihai, Weihai 264209, China; 3. Institute of Optics and Electronics, Chinese Academy of Sciences, Chengdu 610209, China; 4. School of Geosciences and Info-Physics, Central South University, Changsha 410083, China

Abstract:To meet the real-time demand of image restoration for recognition and tracking system, an optimization research on two-dimensional FFT of small size realized in graphics processor unit(GPU) efficiently was done. An analysis of two-dimensional FFT algorithm was analyzed first. And according to the characteristics of GPU, a parallel execution model based on graphics processor was proposed. Based on this model, the optimization research was done considering the aspects of algorithm complexity, the number of jump instructions, access conflict and access latency of the shared memory, and the utilization efficiency of GPU. And two-dimensional FFT computation of small size was realized in the GPU. In image restoration experiment, comparison on the calculation accuracy of two-dimensional FFT of small size optimization algorithm based on GPU and the traditional algorithm in MATLAB based on CPU was done. And a comparison on the computational efficiency of optimization algorithm proposed and the library function image restoration algorithm of CUFFT provided by NVIDIA Corp in four different sizes based on the same GPU platform was made. The results indicate that this optimization algorithm has excellent recovery performance, and human vision system could not distinguish the difference between the results and the MATLAB demonstrations. And the optimization algorithm can recover a frame of 128×128 gray fuzzy image within 19.6 ms, while the computing speed increases 1.8 times approximately compared with that using library function of CUFFT directly.

 

Key words: graphic processing unit (GPU); FFT of small size; image restoration; parallel optimization; real-time computation

中南大学学报(自然科学版)
  ISSN 1672-7207
CN 43-1426/N
ZDXZAC
中南大学学报(英文版)
  ISSN 2095-2899
CN 43-1516/TB
JCSTFT
版权所有:《中南大学学报(自然科学版、英文版)》编辑部
地 址:湖南省长沙市中南大学 邮编: 410083
电 话: 0731-88879765(中) 88836963(英) 传真: 0731-88877727
电子邮箱:zngdxb@csu.edu.cn 湘ICP备09001153号