The Development of Three Image Quality Evaluation Metrics Based on a Comprehensive Dataset

Abstract

In this study, a large scale experiment was carried out to assess the image quality of 2266 images using categorical judgement method by 20 observers. These images were rendered in color contrast, chroma, colorfulness, lightness, and vividness directions. The results were used to derive three No-Reference (NR) Image Quality Estimation Models (IQEMs). The first model was based on color science, (different scales in CIELAB). The second model was a Neural Network model while the third model was a statistics model based on color appearance attributes. Their performances were evaluated using two databases, those developed at Zhejiang University and those available from the public databases in terms of correlation coefficients between the objective and predicted image quality scores.