Resolution Normalized Video Quality Metric
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video quality metrics, such as PSNR and complexity-motion normalized QP, fail to accurately account for distortion due to resampling, leading to inconsistent subjective quality across different video resolutions, especially in lossy encoding and adaptive streaming scenarios.
Innovation Solution
A video processing system that calculates a resolution normalized video quality metric by using a mapping between up-sampled peak signal-to-noise ratio (uPSNR) and a video quality metric (M), allowing for accurate quality assessment across varying resolutions and content types, and is applicable in real-time applications like live HTTP-based adaptive streaming.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional video quality metrics (PSNR, QP) are used, then computation is simple and fast, but measurement precision is poor due to inability to account for resampling distortion
Solution Approach 1:
The system pre-computes and stores mapping tables between PSNR values and quality metric M values for different resolutions during an offline calibration phase. This preliminary action allows the runtime system to simply look up pre-computed values rather than performing complex real-time calculations, thereby improving measurement precision without significantly increasing computational complexity during actual video quality assessment
Solution Approach 2:
The patent introduces an intermediary mapping relationship between PSNR and quality metric M that accounts for resampling distortion. Instead of directly using PSNR or QP as the quality metric, the system uses this intermediate mapping table (which incorporates resolution normalization) to translate simple PSNR calculations into more accurate quality assessments, effectively bridging the gap between computational simplicity and measurement precision
2Measurement precision
If resolution-specific quality metrics are calculated for each video resolution, then measurement precision improves, but loss of time increases due to multiple separate calculations
Solution Approach 1:
The system performs the complex task of calculating quality metrics for multiple resolutions in advance, storing the results in pre-computed mapping tables. During runtime, regardless of the input video resolution, the system simply performs a lookup in the appropriate pre-computed table, reducing the time complexity from O(n) separate calculations to O(1) lookup operation while maintaining precision across all resolutions
Solution Approach 2:
The patent creates a universal mapping table structure that can serve multiple resolutions simultaneously. The mapping between PSNR and quality metric M is established in a resolution-agnostic manner, allowing a single lookup operation to provide accurate quality assessment for any resolution, thereby eliminating the need for separate calculation processes for each resolution and significantly reducing time loss
Data Source
AI summary
In one embodiment, a video processing system includes a processor to read a mapping between a peak signal to noise ratio, PSNR, and a video quality metric, M, for an encoded version of a first video portion encoded at full resolution Rmax, read an up-sampled peak signal to noise ratio, uPSNR(Rx) of an up-sampled decoded video portion deriving from the first video portion being encoded at a resolution Rx, which is less than the full resolution Rmax, yielding an encoded video portion, the encoded video portion being decoded and then up-sampled to the full resolution Rmax; and lookup a value, V′, that maps to uPSNR(Rx) in the mapping, V′ representing a resolution normalized M value of the encoded video portion encoded at the resolution Rx. Related apparatus and methods are also described.