Compressed Video Quality Metric Based on Just-Noticeable-Difference and Saliency-aware Blocking Detection

Abstract

Block-based prediction-quantization hybrid coding framework is widely used in video compression, which results in visually annoying artifacts, i.e., blocking artifacts, on encoded videos. They significantly degrade the Quality-of-Experience (QoE). To improve the QoE, it is crucial to develop Video Quality Assessment (VQA) methods that can identify and quantify various artifacts, among which the negative effect introduced by blocking artifacts is of guiding significance. To evaluate and monitor video quality effectively, a novel Blocking-based Compressed Video Quality Assessment (BCVQA) is proposed in this paper. The key idea of the BCVQA method is a blocking detection model based on Just-Noticeable-Difference (JND) mechanism, which is aimed to improve detection performance, since it only focuses on the observable blocking artifacts. Furthermore, the saliency model is utilized in quality prediction procedure after blocking detection, considering the visual attention mechanism. With the fusion of the above two visual mechanisms, quality prediction of compressed videos becomes more effective, since the useful information is refined in our framework. Experimental results show that the proposed BCVQA is consistent with perceived quality and can effectively evaluate video quality.