基于视觉对比敏感度与恰可察觉失真感知的图像复原  

Image restoration based on visual contrast sensitivity and just noticeable distortion perception

作　　者：杨亚威[1,2] 李俊山[1] 张士杰 芦鸿雁 胡双演[1] 

机构地区：[1]第二炮兵工程大学信息工程系 [2]96215部队 [3]武警工程大学

出　　处：《光学精密工程》2014年第2期459-466,共8页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.61175120)

摘　　要：针对传统复原算法在低信噪比情况下复原结果感知效果差的问题,提出了一种基于视觉对比敏感度与恰可察觉失真感知的图像复原方法。首先,利用离散余弦变换(DCT)分解图像,将图像编码为不同尺度下的一系列子带,利用概率和模型及DCT感知模型计算每条子带的恰可察觉失真阈值,并采用明可夫斯基距离计算失真度量;然后,对于每个尺度下的每条子带,利用一个迭代的两断对分程序来调整每个尺度下的每条子带的通道权重,直至子带的失真度量恰好等于期望的失真阈值;最后,利用对比敏感函数对每条子带中的系数进行视觉敏感度加权,通过DCT逆变换对视觉感知加权后的系数进行重构以得到最终的复原图像。基于LIVE图像库的实验结果表明,本文复原算法不仅能够有效地增加图像的信噪比,而且能够很好地改善图像的感知质量。As traditional restoration algorithms have weaker perceptual opinion in a low Peak Signal Noise Ratio（PSNR）,a novel image restoration approach based on visual contrast sensitivity and Just Noticeable Distortion（JND） perception was proposed.Firstly,the image was coded as a sets of bands in different scales with Discrete Cosine Transform（DCT）,the JND threshold of each band was calculated with probability summation and DCT perception model,and distortion measure was calculated with a Minkowski metric.Then,an iterative bisection procedure was used to adjust the channel weight for each band in each scale until the perceptual distortion of band was just equal to the anticipant distortion threshold.Finally,the coefficient of each band was weighted with visual contrast sensitivity function,and the coefficient after visual perceptual weighting was reconstructed through an inverse DCT to gain the restoration image.Experimental results based on LIVE image database show that this algorithm can effectively enhance PSNR and improve perceptual quality of the recovery images.

关 键 词：图像复原 视觉对比敏感度 恰可察觉失真 离散余弦变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]