No-Reference Video Quality Assessment Using Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video quality assessment methods require significant network bandwidth for full-reference approaches and do not accurately conform to the human visual system, leading to sub-par viewing experiences, especially during peak load scenarios.
Innovation Solution
A system that identifies erroneous videos using visual salience extraction, scene statistics, isolation forest-based anomaly detection, nearest neighbor-based anomaly detection, video segmentation, and normalization, followed by regression modeling to provide accurate no-reference video quality assessment without requiring reference video characteristics transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full-reference approach is used for video quality assessment, then measurement accuracy is improved, but network bandwidth consumption increases
Solution Approach 1:
The patent extracts only the essential characteristics information from the reference video that is necessary for quality assessment, rather than transmitting all reference video data. This extraction approach maintains measurement accuracy while significantly reducing network bandwidth consumption by sending only the minimal required information to the endpoint.
Solution Approach 2:
The patent segments the video quality assessment process into reference video analysis (performed at the server side) and quality evaluation (performed at the endpoint). By dividing the task and performing computationally intensive operations on the server side, the patent reduces the bandwidth requirements at the network transmission level while maintaining full-reference assessment accuracy.
2Loss of energy
If reduced-reference approach is used for video quality assessment, then network bandwidth consumption is reduced, but measurement accuracy deteriorates
Solution Approach 1:
The patent introduces an intermediary processing stage where reference video characteristics are analyzed and transformed into compact representations before transmission. This intermediary step enables the system to maintain full-reference assessment capabilities while using reduced bandwidth, as the intermediary processing extracts and encodes only the essential quality-relevant information.
3Ease of operation
If conventional video quality measurement models are used, then ease of operation is improved, but conformity to human visual system deteriorates
Solution Approach 1:
The patent changes the parameters used in video quality assessment to better align with human visual system characteristics. By modifying the assessment parameters to reflect human perception metrics rather than purely technical video parameters, the patent improves conformity to HVS while maintaining ease of operation through automated processing.
Data Source
AI summary
System and method for identifying erroneous videos and assessing video quality is provided. Feature vectors are generated corresponding to a plurality of frames associated with the one or more videos. The feature vectors are subsequently subjected to anomaly detection to obtain first and second normalized path lengths and normalized anomaly measures. The first and second normalized path lengths and normalized anomaly measures are provided to a regression model to identify the erroneous video.