Video Quality Prediction Using Support Vector Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing objective video quality measurement systems face challenges in efficiently measuring perceptual quality, particularly at endpoints like mobile devices, due to high network bandwidth consumption in full-reference approaches and difficulties in modeling the human visual system effectively.
Innovation Solution
The use of support vector machines (SVMs) to build video quality models based on features from target and reference videos, enabling reduced-reference and no-reference approaches that accurately predict perceptual quality by conforming closer to the human visual system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full-reference approach is used to measure video quality by transmitting all reference video information to the endpoint, then measurement accuracy is improved, but network bandwidth consumption increases excessively
Solution Approach 1:
The patent extracts only the essential reference video information needed for quality assessment, separating the critical features (such as PSNR, SSIM metrics) from the complete reference video data. This allows the endpoint to perform accurate quality measurement without transmitting the entire reference video, thus resolving the contradiction between measurement accuracy and bandwidth consumption.
Solution Approach 2:
The patent segments the video quality assessment process into two parts: feature extraction at the server side and quality prediction at the endpoint. By dividing the reference video into key frames and extracting salient features, the system reduces the data transmission requirement while maintaining measurement accuracy through distributed processing.
2Measurement precision
If prior attempts are made to build video quality models conforming to the human visual system, then measurement relevance to human perception is improved, but model building difficulty increases due to the complex and poorly understood nature of the HVS
Solution Approach 1:
The patent introduces an intermediary quality prediction model that bridges the gap between objective metrics and human perception. Instead of directly modeling the complex HVS, the system uses intermediate representations (such as decoded video frames, feature descriptors) that can be processed by machine learning models to predict perceptual quality, thus simplifying the overall system while maintaining relevance to human vision.
Solution Approach 2:
The patent transforms the complex HVS into a set of manageable parameters and features that can be captured through automated extraction. By changing the representation from direct HVS modeling to parameter-based feature extraction (such as spatial frequency, contrast, color distribution), the system reduces model complexity while maintaining perceptual quality assessment accuracy.
Data Source
AI summary
Systems and methods of objective video quality measurement based on support vector machines. The video quality measurement systems can obtain information pertaining to features of a target training video, obtain corresponding information pertaining to features of a reference version of the target training video, and employ the target training features and/or the reference training features to build video quality models using such support vector machines. Based on the target training features and/or the reference training features used to build such video quality models, the video quality models can be made to conform more closely to the human visual system. Moreover, using such video quality models in conjunction with target features of a target video whose perceptual quality is to be measured, and/or reference features of a reference video, the video quality measurement systems can be employed to predict measurements of the perceptual quality of such a target video with increased accuracy.