基于非负矩阵分解的彩色图像质量评价方法

针对稀疏表示的图像质量评价模型都基于灰度图像，缺少颜色信息，该文提出一种基于非负矩阵分解(NMF)的全参考彩色图像质量评价方法。首先，从自然彩色图像中随机采样，得到训练样本，利用非负矩阵分解，训练得到特征基矩阵，并经过Schmidt正交化，构建特征提取矩阵；其次，根据视觉显著性模型，利用最大视觉显著性和显著性差值两步骤选取视觉重要区域；最后，利用特征提取矩阵，得到低维的特征向量，并最终得到彩色图像质量评价值。实验结果表明，该文方法在LIVE, CSIQ和TID2008 3个图像质量评价库上有很好的表现。3个图像库的平均结果显示，该文方法的综合表现优于所有对比方法。这表明该文方法与主观感知有更好的关联度。

A Color Image Quality Assessment MethodBased on Non-negative Matrix Factorization

For the sparse representation of image quality assessment model are based on gray image and the lack of color information, a Non-negative Matrix Factorization (NMF)-based full reference color image quality assessment method is proposed. Firstly, from the natural color image in random sampling, training samples are got. Non-negative matrix factorization method is used to train and get a feature basis matrix. After using Schmidt orthogonalization, a feature extracting matrix is got. Secondly, according to the visual saliency model, maximum visual saliency is defined and significant difference of two steps is used to select visual important area. Finally, using the feature extraction matrix, low dimensional feature vectors and the final color image quality evaluation value are got. The experimental results show that the proposed method has good performance in the LIVE, CSIQ and TID2008 three image databases. The average results of three image quality assessment databases show that the proposed method outperforms other methods, which means that the proposed method has better correlation with the subjective perception.