Methods, systems, and media for determining perceptual quality indicators of video content items
Topics: AI (Deep Learning), Ranking, Semantic Search, Video Search
This Google patent describes a system and method for automatically assessing video quality using deep neural networks. The technology analyzes video content frame-by-frame using three specialized neural network components that evaluate content quality, distortion, and compression levels. The system generates both per-frame quality scores and an overall video quality assessment, which can be used to optimize video compression, provide feedback to content uploaders, and improve streaming service quality.