为了设计与人的主观评价相吻合的全参考型客观图像质量评价(IQA)算法。针对不同算法提取的局部特征, 利用广义平均的非线性性质, 提出了2种池化策略, 以提高结构相似度(SSIM), 梯度结构相似度 (GSSIM), 特征相似度指标 (FSIM)的评价能力。在TID2008和TID2013数据库中进行数值实验, 讨论了所有失真类型非线性参数的选择以及不同失真类型之间非线性参数的变化。结果表明, 采用广义平均池化策略能提高IQA算法的有效性。4种客观评价指标Spearman等级相关系数(SROCC)、Kendall等级相关系数(KROCC)、Pearson线性相关系数(PLCC)和均方误差根(RMSE)表明所提算法性能优于已有的算法, 与人的视觉系统具有一致性。

In order to design a full-reference objective Image Quality Assessment (IQA) algorithm that consistent with subjective evaluation. Based on local feature extracted according to different algorithms and nonlinear properties of generalized means strategy, two pooling strategies were proposed to promote the ability to evaluate Structural Similarity Image Measurement (SSIM), Gradient Structural Similarity Image Measurement (GSSIM) and Feature Similarity Index (FSIM). Numerical test was conducted in TID2008 and TID2013 database, selections of various distortion non-linear parameters as well as the changes of non-linear parameters among different distortion types were discussed. The results show that the application of general means strategies could promote the effectiveness of IQA algorithm. 4 kinds of objective evaluation indexes, which are Spearmans Rank-Order Correlation Coefficient (SROCC), Kendalls Rank-Order Correlation Coefficient (KROCC), Pearsons Linear Correlation Coefficient (PLCC) and the Root Mean Square Error (RMSE), indicate that the algorithm proposed herein is superior to the existing algorithm, proves the consistency with human visual system.