A No-reference Image Quality Assessment based on Reference Generating Network
Sarala Ghimire, Nagaraj Yamanakkanavar, Bumshik Lee · 2020
In practice, owing to the lack of original information, no-reference image quality assessment (NR-IQA) that does not require a reference image is more attractive. However, the quality predicted by the full reference IQA that utilizes the full information of reference image is highly correlated with the subjective quality score and thus the prediction accuracy is higher than the NR-IQA. Hence to alleviate the accuracy difference and to imitate the FR-IQA performance, we proposed the NR-IQA method using a reference generating network, where the reference image is reconstructed, and relative associativity with the distorted image is calculated to find the final quality score. The proposed architecture is significantly simple and efficient. In addition, the comprehensive experimental result shows that the prediction accuracy is comparable with that of the state-of-the-art methods.