视觉掩盖效应下的无参考图像信噪比评价  

No-reference image SNR assessment under visual masking

作　　者：范媛媛[1] 桑英军[1] 沈湘衡[2] 

机构地区：[1]淮阴工学院,江苏淮安223003 [2]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《应用光学》2012年第4期711-716,共6页Journal of Applied Optics

基　　金：中国科学院创新基金(YZ200904);淮阴工学院青年教师基金(HGC1112)

摘　　要：针对噪声图像,提出一种视觉掩盖效应下的无参考图像信噪比评价方法。采用Hosaka分块思想,利用四叉树分解,取消了Hosaka分块对图像尺寸的限制,分块的大小体现出块内图像的主要频率成分;然后检测各个子块噪声,根据图像的污染程度,计算NRPSNR值。利用LIVE Database Release2图像库中的WN的174幅图像检验NRPSNR与DMOS的相关度,并与MSE、PSNR以及HVSNR比较分析。实验结果表明,该方法总体评价效果优于其他3种评价方法,CC为-0.895 0,SROCC为-0.923 6,OR为0.042 9,主客观评价比较吻合,且对于其他噪声类型同样适用,体现了人的视觉掩盖效应,可以满足对无参考噪声图像进行准确可靠的质量评价要求。A novel no reference image signal to noise ratio（SNR） assessment method based on the property of human visual system （HVS） was proposed to evaluate the quality of noisy im- age. Firstly, the method adopted Hosaka plots idea and quadtree decomposition which wiped out the limits on image size. Secondly, this improved method was applied to the image to make plots and was employed to abstract sub-blocks noises. Then, the no reference image signal to noise ratio （NRPSNR） was computed according to the pollution degree. The method was test- ed and verified by using the LIVE Database Release2 image database. The experiment results show that this no reference assessment method which CC reaches -0. 895 0, SROCC reaches -0. 923 6, OR reaches 0. 042 9,is superior to other three kinds of methods on the whole, it is more consistent with subjective image quality assessment, suitable for other noisy styles and embodies the HVS visual masking effect. It can give accurate and reliable quality assessment of no reference noisy images.

关 键 词：无参考图像质量评价 视觉掩盖 人眼视觉系统 信噪比 四叉树分解 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]