基于解析稀疏表示的图像模糊无参考快速评价算法

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 approach based on analysis sparse representation.With the consideration that blur caused loss of high-frequency components in images, the proposed method evaluated blur by estimating energy in an image.Firstly,it decomposed the blurred image using an analy-sis dictionary,and then computed the energy of the image using the sparse coefficients.It used the energy of an image to mea-sure the extent of blur.To reduce the influence of varying image contents,it used the summed variances of image blocks to normalize the energy.Furthermore,it incorporated visual saliency in the pooling so as to produce blur scores that were consis-tent with subjective evaluations.The proposed method tested its performance on four public image quality databases.Experi-mental results demonstrate that the proposed method outperforms the state-of-the-arts with low computational complexity.