融合全局-局部特征的双通道无参考图像质量评价算法研究

Research on Dual Channel Non Reference Image Quality Evaluation Combined with Global Local Features

针对目前大多数图像质量评价(Image Quality Assessment,IQA)算法在对非均匀失真图像进行质量评估时效果不佳的问题,提出一种结合全局-局部特征的双通道无参考图像质量评价(No-Reference Image Quality Assessment,NR-IQA)算法。首先,考虑输入图像尺寸的不同,利用局部失真重组算法对输入图像进行预处理。其次,利用基于Swin Transformer模块的双通道神经网络提取图像的全局特征和局部特征。最后,通过质量回归预测网络完成全局-局部特征到图像质量分数的映射。实验结果表明,该算法在两个数据集上分别取得0.823和0.871的斯皮尔曼等级相关系数(Spearman Rank Order Correlation Coefficient,SROCC)指标值,表明所提出算法与人的主观感知较为吻合。

Aiming at the problem that most Image Quality Assessment(IQA)algorithms are not effective in evaluating the quality of non-uniform distorted images,A two-channel No-Reference Image Quality Assessment(NR-IQA)algorithm combining global and local features is proposed.Firstly,considering the different sizes of input images,the local distortion recombination algorithm is used to preprocess the input images.Secondly,the two-channel neural network based on Swin Transformer module is used to extract the global and local features of the image.Finally,the global-local feature is mapped to the image quality score through the quality regression prediction network.The experimental results show that the Spearman Rank Order Correlation Coefficient(SROCC)index values of 0.823 and 0.871 are obtained on the two datasets,respectively,indicating that the proposed algorithm is in good agreement with human subjective perception.