Bitstream-Based Video Quality Models for Encoding Profile Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Determining optimal encoding parameters for streaming video services with diverse content characteristics is challenging, especially with codecs having numerous parameters, and existing subjective quality assessment methods are expensive and impractical for live streaming.
Innovation Solution
A bitstream-based video quality assessment (VQA) model using machine learning to predict quality scores without decoding, incorporating features like quantization parameters, motion vectors, and bitstream metadata to improve accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If subjective viewer assessments are used to determine video quality, then measurement precision is improved, but loss of time and productivity deteriorate due to expensive and complex testing processes
Solution Approach 1:
The patent creates an objective VQA model that copies the human visual system's response characteristics to predict perceived video quality without requiring actual human viewers. The model uses machine learning algorithms trained on subjective quality data to replicate human quality assessment, enabling automated quality prediction that maintains measurement precision while dramatically improving productivity.
Solution Approach 2:
The patent replaces the mechanical system of human viewer assessment with an automated computational system. Instead of relying on human subjects to evaluate video quality, the system uses machine learning models that process video features and bitstream data to predict quality scores, eliminating the time-consuming and resource-intensive nature of subjective testing.
2Manufacturing precision
If numerous encoding parameters are used in video codecs, then manufacturing precision is improved for video quality optimization, but device complexity increases with 30-50 parameters to manage
Solution Approach 1:
The patent changes the approach to parameter management by using machine learning models that automatically learn optimal relationships between multiple encoding parameters and video quality. Instead of manually managing 30-50 parameters, the system trains models on encoded video data to automatically determine which parameter combinations yield best quality, transforming complex parameter management into an automated optimization process.
Solution Approach 2:
The patent enables the encoding system to self-optimize by using VQA models to automatically evaluate the quality impact of different parameter settings. The system performs self-assessment of encoding quality and automatically adjusts parameters based on predicted quality outcomes, eliminating the need for manual parameter tuning and reducing operational complexity.
3Loss of time
If bitstream-based VQA model is used instead of pixel domain analysis, then loss of time is reduced for quality assessment, but measurement precision may deteriorate without access to decoded video frames
Solution Approach 1:
The patent transitions from analyzing video in the spatial pixel domain to analyzing features in the bitstream domain. By extracting quality-relevant features directly from the encoded bitstream data rather than from decoded pixel values, the system achieves fast quality assessment without sacrificing measurement precision. This dimensional shift allows quality evaluation to occur at the encoding stage without requiring full video decoding.
Data Source
AI summary
Techniques are described for training and use of machine learning models to determine objective video quality scores. Video quality scores predict the quality of video content perceived by viewers. Quality scores have various uses, including the selection of encoding profiles and determination of encoding ladders. A core model and residual model may be used to determine quality scores.