Using Spatio-Temporal Saliency to Predict Subjective Video Quality: A New High-Speed Objective Assessment Metric

TLDR: A new VQA metric, consisting of a method based on spatio-temporal saliency to model human visual perception of quality, called Sencogi Spatio-Temporal Saliency Metric (Sencogi-STSM), which generates subjective quality scores of video compression in terms of prediction efficacy and accuracy more than the most used objective V QA models.

Abstract: We describe a new Objective Video Quality Assessment (VQA) metric, consisting of a method based on spatio-temporal saliency to model human visual perception of quality. Accurate measurement of video quality is an important step in many video-based applications. Algorithms that are able to significantly predict human perception of video quality are still needed to evaluate video processing models, in order to overcome the high cost and time requirement for large-scale subjective evaluations. Objective quality assessment methods are used for several applications, such as monitoring video quality in quality control systems, benchmarking video compression algorithms, and optimizing video processing and transmission systems. Objective Video Quality Assessment (VQA) methods attempt to predict an average of human perception of video quality. Therefore subjective tests are used as a benchmark for evaluating the performance of objective models. This paper presents a new VQA metric, called Sencogi Spatio-Temporal Saliency Metric (Sencogi-STSM). This metric generates subjective quality scores of video compression in terms of prediction efficacy and accuracy than the most used objective VQA models. The paper describes the spatio-temporal model behind the proposed metric, the evaluation of its performance at predicting subjective scores, and the comparison with the most used objective VQA metrics.