提升小波变换在NAS-RIF盲复原算法中的应用  

Application of Lifting Wavelet Transform in Blind Restoration Scheme Based on NAS-RIF Algorithm

作　　者：黄德天[1,2] 吴志勇[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所光电测控部,长春130033 [2]中国科学院研究生院,北京100049

出　　处：《计算机辅助设计与图形学学报》2012年第12期1614-1620,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家"八六三"高技术研究发展计划(2008AA7034320B)

摘　　要：针对采用非负支撑域受限递归逆滤波(NAS-RIF)算法对低信噪比图像进行复原时将导致算法性能恶化的问题,提出一种与提升小波变换相结合的NAS-RIF盲复原算法.首先对退化图像进行整数提升小波分解,得到不同频带子图像的信息;然后对各个频带子图像分别采用基于空间自适应和正则化方法的NAS-RIF算法进行复原,针对不同频带子图像的频率和方向特性,通过自适应地选取对应的正则化算子、正则化参数和空域加权因子,达到对低频子图像去模糊、对高频子图像抑制噪声,并保持边缘细节的目的;最后通过整数提升小波逆变换得到复原后的图像.在不同的信噪比条件下对2种模糊图像进行仿真实验,采用文中算法得到的信噪比增益分别为5.849 1dB和9.713 6dB.实验结果表明,文中算法不仅取得了更优的图像复原效果,而且具有较快的收敛速度.To solve the problem that the performance of the non-negativity and support constraint recursive inverse filtering （NAS-RIF） algorithm decreases while it is applied to restore degraded images under a low signal noise ratio （SNR） condition, a NAS-RIF algorithm based on lifting wavelet transform for blind restoration is proposed. Firstly, the degraded image is decomposed by integer lifting wavelet transform, and its wavelet coefficients in wavelet domain are obtained in different frequency sub-bands. Secondly, the improved NAS-RIF algorithm based on space-adaptive and regularization is employed to restore degraded image in each sub-bands. According to different frequency and orientation characteristic in different sub-bands, corresponding regularization operators, regularization parameters and space-weights are adaptively selected to remove blurs in low frequency sub-bands, reduce noise and preserve edges in high frequency sub-bands. Finally, the restored image is reconstructed by the inverse integer lifting wavelet transform. The simulating experiments on two kinds of degraded images are performed under different SNR conditions, and the ASNRs by proposed algorithm are 5. 849 1 clB and 9. 713 6 dB, respectively. The experiment results demonstrate that the proposed algorithm not only obtains better restoration results, but has faster convergence rate.

关 键 词：图像盲复原 非负支撑域受限递归逆滤波算法 提升小波变换 空间自适应 正则化方法 N步重置共轭梯 度法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]