基于多阶结构表示的超分辨率重构图像质量盲评价方法

Blind quality metric of super-resolution reconstructed images based on multi-order structure representation

为了客观地预测超分辨率重构图像的质量,提出了一种基于多阶结构表示的超分辨率重构图像无参考质量评价方法。利用多阶导数信息表示超分辨率重构图像的主要结构和细微纹理,并利用局部二值模式提取超分辨率重构图像的多阶结构特征;结合主观分数,利用随机森林回归训练图像质量预测模型,再利用模型预测待测图像的质量。为了证明该算法的有效性和优越性,对比实验在一个大尺度超分辨率图像数据库上进行。该算法的斯皮尔曼相关系数和皮尔森线性相关系数分别为0.910 3和0.918 3。实验结果表明,该算法优于现有的无参考质量评价算法,与主观评价结果保持较高的一致性。此外,该算法时间复杂度低,运行时间适中。

In order to objectively predict the quality of super-resolution(SR)reconstructed images,this paper presents a blind quality metric for SR reconstructed images based on multi-order structural representation.Firstly,the dominant structures and minor textures of a SR reconstructed image are represented by multi-order derivative information,and the quality-aware multi-order structural features are extracted by local binary pattern.Then,combined with subjective scores,the random forest is used to train the image quality prediction model.Finally,the model is used to predict the quality of the image.To prove the effectiveness and superiority of the proposed metric,the comparison experiments are carried out on a large-scale SR reconstructed image database.The spearman’s rank ordered correlation coefficient and pearson linear correlation coefficient of our metric are 0.9103 and 0.9183,respectively.The experimental results show that the proposed metric is superior to the existing quality metrics and maintains high consistency with the subjective evaluation results.In addition,the proposed metric has low time complexity and moderate running time.