No-Reference Video Quality Assessment Using Multi-Pooled, Saliency Weighted Deep Features and Decision Fusion

With the constantly growing popularity of video-based services and applications, no-reference video quality assessment (NR-VQA) has become a very hot research topic. Over the years, many different approaches have been introduced in the literature to evaluate the perceptual quality of digital videos. Due to the advent of large benchmark video quality assessment databases, deep learning has attracted a significant amount of attention in this field in recent years. This paper presents a novel, innovative deep learning-based approach for NR-VQA that relies on a set of in parallel pre-trained convolutional neural networks (CNN) to characterize versatitely the potential image and video distortions. Specifically, temporally pooled and saliency weighted video-level deep features are extracted with the help of a set of pre-trained CNNs and mapped onto perceptual quality scores independently from each other. Finally, the quality scores coming from the different regressors are fused together to obtain the perceptual quality of a given video sequence. Extensive experiments demonstrate that the proposed method sets a new state-of-the-art on two large benchmark video quality assessment databases with authentic distortions. Moreover, the presented results underline that the decision fusion of multiple deep architectures can significantly benefit NR-VQA.