Multi-Feature 360 Video Quality Estimation

We propose a new method for the visual quality assessment of 360-degree (omnidirectional) videos. The proposed is based on computing multiple spatio-temporal objective features viewports extracted from A model learnt to properly combine these into metric that closely matches subjective scores. main motivations approach are that: 1) metrics computed better captures user experience than projection domain; 2) use easily supports different methods being used in current video systems; and 3) no individual image always performs best all types distortions, while learned combination them able adapt conditions. Experimental results, both largest available videos dataset cross-dataset validation, demonstrate outperforms state-of-the-art 2D metrics.