Is It Possible to Predict Human Perception of Video Quality? The Assessment of Sencogi Quality Metric

Sencogi Quality Metric (SenQM) is a novel objective metric for video quality assessment. SenQM infers video quality scores from the spatio-temporal evolution of videos. The quality model behind SenQM is based on an algorithm developed by Cogisen for modelling dynamic phenomena generated by complex systems. In the field of video compression, Cogisen’s algorithm uses machine learning to model human perception of video quality by extracting meaningful information directly from the video data domain and its frequency representation. The model has been trained over datasets of (i) x264 compressed videos as input data and (ii) the corresponding subjective Mean Opinion Scores as ground truth. This study introduces the model behind SenQM and how the proposed metric performs in subjective video quality prediction compared to the most used video quality assessment methods, i.e. PSNR, SSIM, and Netflix’s VMAF. Results indicate a significantly higher prediction performance in terms of monotonicity, consistency, and accuracy than the compared metrics. SenQM quality scores show significantly higher variations for 352 × 288 resolution videos with equivalent levels of degradation, and outstands PSNR, SSIM, and VMAF in predicting subjective scores of increasing levels of compression without being affected by either the degradation level or the video content.