Spatiotemporal Masking for Objective Video Quality Assessment.

Random background and object motion may mask some distortions in video sequence, the masked distortions are ignored by humans and they aren’t considered when humans assess video quality. The visual masking effect produces a gap between the subjective quality and predicted quality obtained by traditional video quality assessment (VQA) which measures all distortions to predict video quality. This paper proposed a novel spatiotemporal masking model (STMM) consists of spatial and temporal masking coefficients to narrow the gap. The spatial masking coefficient is computed by spatial randomness to count the error score between the subjective and objective score, and the temporal masking coefficient is combined by three parts that fused by eccentricity, magnitude of motion vectors and coherency of object motion to measure the degree of the masking effect. In addition, the proposed model is robust enough to integrate with several best known VQA metrics in the literature. The improvement achieved by utilizing the proposed model is evaluated in the LIVE database, MCL-V database and IVPL database. Experimental results show that the VQA metric based on STMM has a good consistency with the subjective perception and performs better than its original metric.