基于稀疏表示与能量分解的无参考图像质量评价

No-reference image quality assessment based on sparse representation and energy splitting

针对基于稀疏表示的图像质量评价算法特征信息挖掘不充分、忽略稀疏特性等问题进行了研究,提出了一种基于稀疏表示与能量分解的无参考图像质量评价方法。首先利用L1范数作为稀疏惩罚项学习稀疏编码字典,并计算待评价图像的稀疏表示系数;然后对稀疏系数矩阵进行奇异值分解,并重建若干个等能量的子矩阵;最后联合max-pooling和L1范数描述稀疏系数矩阵及其子矩阵特征,L1范数刻画了稀疏性,子矩阵丰富了特征信息。实验结果表明,该算法能在无参考的情况下更好地评价图像质量,主客观分值一致性好,且时间复杂度较低,具有较好的应用价值。

Focusing on the problems of the existing sparse representation based image quality assessment methods which has poorly mining features and ignoring sparsity,this paper proposed a new no-reference image quality assessment method based on sparse representation and energy splitting(SRAES).Firstly,it took L1-norm as sparse penalty term in the process of learning dictionary of sparse coding and computing sparse representation.Secondly,it used SVD to decompose sparse coefficients and reconstructed several equal-energy sub-matrices.Finally,it united max-pooling and L1-norm to describe the features of sparse coefficients matrix and its sub-matrices,where L1-norm depicted sparsity and these sub-matrices enriched the feature information.The experimental results show that this method yields superior performance with high consistency and low computational complexity,and it is suitable for applications.