基于小波变换的图像Wiener滤波并行实现  

Image Wiener filtering Parallel Realization Based on Wavelet Transform

作　　者：钟升[1] 杨恒[2] 王忠[3] 

机构地区：[1]西安微电子技术研究所,西安710065 [2]西安通信学院军事电子工程系,西安710106 [3]第二炮兵工程学院,西安710025

出　　处：《信号处理》2008年第2期333-338,共6页Journal of Signal Processing

基　　金：国防微电子预研项目(41308010203)

摘　　要：面向大规模数据图像去噪处理高效实现,本文提出一种基于Daubechies 9—7小波分解的方向窗Wiener滤波并行实现方法和阵列结构。首先针对小波变换给出了提升小波变换的并行实现方法和PE间的互联结构。该实现方法极大地加速了小波变换的速度,同时其互联结构又确保了各变换层中存放相同子带系数的PE的直接互联,有效地减少了小波变换和去噪处理中的通信开销。针对Wiener滤波中各子带滑动窗累加求和问题,采用循环前缀求和的处理方式,并行地实现了多窗口的累加求和,极大地降低了Wiener滤波的时间开销,满足大图像帧处理的实时性要求。To archieve highly efficient realization of denoising process of large-scale images, a kind of wiener filtering parallel implementing method and array architecture based on danbechies 9-7 bi-orthogonal wavelet transform are proposed in this paper. First, a parallel implementing method of lifting wavelet transformation and the PE interconnect are introduced. While the former accelerates the decomposing process, the latter ensures direct connection between PEs which store the same subband coefficients of every transform level. In this way, the communication cost of wavelet transformation and denoising are efficiently reduced. Second, focusing the accumulation problem of subband sliding window during Wiener filtering, the way of cycling prefix-accumulation is adopted to realize the parallel accumulation of multi-windows, which dramatically reduces time cost of wiener filtering, and real-time processing requirement of large scale image frames are satisfied.

关 键 词：提升小波变换 方向窗Wiener滤波 PM2I网络 前缀求和 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP301.6[自动化与计算机技术—计算机科学与技术]