No-Reference Video Quality Assessment Using Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video quality assessment methods require a reference video of pristine quality, which is often unavailable in real-world video distribution networks, limiting their applicability and effectiveness in monitoring video quality across multiple points in the network.
Innovation Solution
A no-reference (NR) objective video quality assessment (VQA) method that combines deep neural networks (DNNs) with domain knowledge, including models of the human visual system, content analysis, distortion analysis, and viewing device conditions, to produce an overall quality score without relying on a reference video.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full-reference or reduced-reference VQA methods are used, then measurement precision of video quality is improved, but device complexity and requirement for reference video availability worsen
Solution Approach 1:
The patent extracts and removes the reference video requirement from the VQA system, creating a no-reference methodology that achieves acceptable quality assessment without the complex infrastructure of reference video distribution and synchronization
Solution Approach 2:
The patent introduces HVS models and content-aware processing as intermediary components that bridge the gap between simple pixel-based metrics and complex perceptual quality assessment, enabling accurate quality measurement without reference videos
2Ease of operation
If no-reference VQA methods are used, then ease of operation and applicability in video distribution networks is improved, but measurement precision of video quality assessment deteriorates
Solution Approach 1:
The patent transforms the quality assessment approach by changing from reference-based parameters (comparing to pristine video) to no-reference parameters (using HVS models, content analysis, and distortion detection), maintaining precision while improving ease of operation
Solution Approach 2:
The patent combines multiple assessment components (HVS models, content analysis, distortion detection) into a composite no-reference VQA system that achieves precision comparable to full-reference methods while being easier to operate in real-world networks
3Measurement precision
If deep neural networks are used for quality assessment, then measurement precision is improved, but use of energy and computational resources worsens
Solution Approach 1:
The patent segments the quality assessment task into multiple specialized DNNs that process different aspects (spatial quality, temporal quality, content characteristics), allowing selective execution and reduced overall computational energy compared to a single large DNN
Solution Approach 2:
The patent implements content-aware processing that selectively applies full DNN-based assessment only when needed (e.g., when content complexity or distortion levels warrant it), using simpler methods for routine cases to reduce energy consumption
Data Source
AI summary
No-reference (NR) quality assessment (VQA) of a test visual media input encoding media content is provided. The test video visual media input is decomposed into multiple-channel representations. Domain knowledge is obtained by performing content analysis, distortion analysis, human visual system (HVS) modeling, and/or viewing device analysis. The multiple-channel representations are passed into deep neural networks (DNNs) producing DNN outputs. The DNN outputs are combined using domain knowledge to produce an overall quality score of the test visual media input.