基于解析稀疏表示的图像模糊无参考快速评价算法  

Fast no-reference image blur assessment based on analysis sparse representation

作　　者：张薇[1] 蔡浩[1] 胡波[1] 郭星歌[1] Zhang Wei;Cai Hao;Hu Bo;Guo Xingge(School of Information & Electrical Engineering, China University of Mining & Technology, Xuzhou Jiangsu 221116 , China)

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221116

出　　处：《计算机应用研究》2016年第10期3169-3172,3190,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61379143);徐州市科技计划资助项目(XM13B119)

摘　　要：针对模糊图像的质量评价,提出了一种基于解析稀疏表示的图像模糊快速评价算法。考虑到模糊会造成图像中高频信息的损失,因此所提出的方法通过衡量图像中的能量来评价图像的模糊程度。首先利用解析稀疏表示字典将待评价图像进行分解,并通过稀疏表示系数计算图像的能量,以作为图像模糊程度的指标。为了消除图像内容对评价结果的影响,采用图像块的方差和来对图像的能量进行归一化。然后利用视觉显著性进行加权,使得模糊质量分数与主观评价结果更加一致。最后在四个通用的图像质量数据库上对算法的性能进行测试。实验结果表明该算法的性能优于现有算法,且计算复杂度较低。For the quality assessment of blur images, this paper presented a fast no-reference image blur assessment approachbased on analysis sparse representation. W ith the consideration that blur caused loss of high -frequency com ponents in im ages,the proposed m ethod evaluated b lu r by estim ating energy in an im age. Firstly , it decomposed the blur redimage using an an alysisdictionary , and then com puted the energy o f the image using the sparse co efficients. I t used the energy o f an image to measurethe extent o f b lu r. To reduce the in fluence of varyin g image contents, it used the summed variances o f image blocks tonorm alize the energy. F u rth e rm o re , it incorporated visu a l saliency in the pooling so as to produce b lu r scores that were consisten t w ith subjective evaluations. The proposed m ethod tested its perform ance on four public image quality databases. Experimental results dem onstrate that the proposed method outperform s the state-of-the-arts w ith low com putational complexity.

关 键 词：图像质量评价 模糊 无参考 解析稀疏表示 视觉显著性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]