Computer Engineering and Applications ›› 2008, Vol. 44 ›› Issue (2): 120-122.

• 产品、研发、测试 • Previous Articles     Next Articles

Implementation and performance of FFT and convolution in image filtering on GPU

FENG Huang   

  1. Siemens Shanghai Mobile Communications Ltd,Shanghai 200126,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2008-01-11 Published:2008-01-11
  • Contact: FENG Huang

GPU图像处理的FFT和卷积算法及性能分析

冯 煌   

  1. 上海西门子移动通信有限公司,上海 200126
  • 通讯作者: 冯 煌

Abstract: Image filtering approaches are important part of many contemporary image-processing tools.Since image-filtering approaches are very computationally demanding,to preserve interactivity of those tool,using Graphics-Process-Unit(GPU) to accelerate the image filtering may be a good way out.In this article the author gives the GPU implementation of two basic approaches to image filtering Fast Fourier Transform(frequency domain) and convolution(spatial domain),and evaluate these methods in terms of the performance and suitability for GPU implementation.Convolution yields better performance in many cases,and the author also identifies conditions under which the FFT givers better performance.

Key words: Fast Fourier Transformation(FFT), convolution, Graphics Procession Unit(GPU), image filtering

摘要: 图像滤波器是当前绝大多数图像处理软件中的重要组成部分;然而,图像滤波对于计算量的要求是巨大的,为了加强图像处理软件的人机交互性能,使用GPU(可编程图形处理器)来加速图像滤波,是一个很好的选择。讨论了在GPU上两种图像处理工具的实现:频域上的快速傅立叶变换和空间域上的卷积运算,并评估了这两种工具在GPU上的性能表现。卷积运算在一般情况下表现出来比FFT更好的性能;并同时讨论了在FFT运算有更佳性能的情况。

关键词: 快速傅立叶变换, 卷积, 可编程图形处理器, 图像滤波