Distortion based image quality index
Jingwei Guan, Wai-Kuen Cham · 2016
In this paper, we tackle the problem of no-reference image quality assessment. This paper proposes a non-distortion-specific image quality evaluator, i.e., deep learning based blind image quality index DL-BIQI, which trained several deep models to estimate the visual quality. Since different distortion types lead to the different influence on images, each model is designed for a specific distortion type. Meanwhile, another deep classification model is proposed to estimate the presence of a set of distortions in the testing image. The final visual quality is obtained by a probability-weighted summation. Experiments were conducted on the LIVE dataset [1] to evaluate its effectiveness. The performance of the proposed method achieves 0.951 for SROCC. It outperforms the state-of-the-art methods for comparison. Besides, it is shown that the proposed deep classification model achieves 93.7% accuracy on the LIVE dataset.