Reduction of temporal distortion in video coding based on detection of just-noticeable temporal pumping artifact

• A metric of just-noticeable temporal pumping artifact (JNTPA) is formulated.
• The proposed metric predicts JNTPA values in correspondence with human perception.
• An efficient algorithm based on the JNTPA has been devised for TPA reduction.
• The proposed algorithm achieves a more smooth subjective video quality.
• The proposed algorithm improves the viewing comfort for the human eye.

At low bit-rates, video coding under the hierarchical prediction structure induces a temporal distortion known as temporal pumping artifact (TPA) when the quantization parameter cascading strategy is used. TPA manifests itself as a pumping effect visually, due to severe quality fluctuations among adjacent pictures, which seriously affects the perceptual quality of a video. In this work, the causes and perception of the TPA is first analyzed. Then a metric of just-noticeable temporal pumping artifact (JNTPA) is formulated based on characteristics of the human visual system. Based on JNTPA, an efficient algorithm has been devised for TPA reduction (denoted as TRA-JNTPA for short). The experimental results have demonstrated that the proposed JNTPA measure is in line with human perception, and compared with JVT-P014, the proposed TRA-JNTPA achieves significant reduction of the TPA with superior subjective video quality at a given bit-rate.