Search

Most full-reference image quality assessment (IQA) models first compute local quality scores and then pool them into an overall score. In this paper, we develop an innovative pooling strategy based on sample statistics to adaptively make the IQA more consistent with human visual assessment. The innovation of this work is threefold. First, we identify that standard sample statistics and robust sample statistics could provide complementary information about the degree of degradation in distorted images. Second, an effective IQA metric is proposed by adaptively integrating robust sample statistics and standard sample statistics via excess kurtosis. Third, instead of using the statistics directly, we adjust them by taking into account the global change of image gradients to avoid exaggerating the degradation degree. Experiments conducted on five well-known IQA databases demonstrate the effectiveness of the proposed pooling strategy in terms of high prediction accuracy and monotonicity.