Video Quality Assessment via Spatiotemporal Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video quality assessment methods rely on subjective human tests, which are laborious and prone to variability, and struggle with accurately estimating subjective quality due to insufficient physical feature values and inadequate spatial and temporal alignment of reference and deteriorated video signals, especially in diverse video formats and networks.
Innovation Solution
A video quality assessing apparatus and method that calculates and aligns physical feature values of reference and deteriorated video signals, using correction information and alignment processes to estimate subjective quality, incorporating edge energy, moving energy, and temporal information to correct for spatial and temporal deviations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If subjective quality assessing tests with human testers are conducted, then accurate subjective quality assessment can be obtained, but the process becomes laborious and time-consuming
Solution Approach 1:
The patent creates an objective assessment system that copies the human subjective assessment process by training a model on human assessment data. The system learns to predict subjective quality metrics through physical measurements and machine learning, replacing the need for actual human observers while maintaining assessment accuracy
Solution Approach 2:
The patent replaces the mechanical system of human visual inspection with an automated computational system that uses physical measurements, signal processing, and machine learning algorithms to assess video quality objectively, eliminating manual labor and time consumption
2Productivity
If physical feature values are used to estimate subjective quality, then the assessment process becomes automated and faster, but the accuracy is insufficient without proper spatial and temporal alignment
Solution Approach 1:
The patent performs spatial and temporal alignment as preliminary actions before quality assessment. The system aligns reference and deteriorated video signals in both space and time domains, ensuring that comparisons are made between corresponding frames and regions, which is essential for accurate objective assessment
Solution Approach 2:
The patent transforms the assessment approach by changing parameters from simple physical measurements to aligned spatio-temporal feature comparisons. The system adjusts and normalizes various parameters including frame timing, spatial positioning, and feature scaling to enable accurate automated assessment
3Reliability
If conventional alignment techniques are used for TV broadcasting signals, then synchronized state can be maintained, but the method fails for IP network video signals with varying packet arrival intervals and losses
Solution Approach 1:
The patent implements dynamic alignment techniques that adapt to varying network conditions. The system continuously adjusts alignment parameters based on real-time analysis of packet arrival patterns, timing variations, and signal characteristics, enabling reliable synchronization for IP network video signals with fluctuating delivery conditions
Solution Approach 2:
The patent creates a universal alignment framework that works across multiple video formats and transmission protocols. The system handles both traditional TV broadcasting signals and modern IP network video signals, accommodating various packetization schemes, timing mechanisms, and network conditions through a unified approach
4Device complexity
If feature values are insufficient, then the assessment method is simpler, but the accuracy cannot match human assessment
Solution Approach 1:
The patent combines multiple types of features and measurements into a composite assessment model. The system integrates spatial features, temporal features, statistical measures, and machine learning predictions to create a comprehensive quality assessment that achieves human-level accuracy through the synergistic combination of multiple indicators
Data Source
AI summary
A subjective quality estimating part (11) receives an undeteriorated reference video signal (RI) and a deteriorated video signal (PI) produced from the reference video signal, calculates video signal feature values for both the signals, and according to a difference between the calculated video signal feature values of the signals, estimates a subjective quality of the deteriorated video signal. A feature value calculating part (12) calculates the video signal feature values of the reference video signal. A correction information storing part (13) stores correction information that corresponds to video signal feature values and is used to correct the subjective quality. A correction calculating part (14) receives the video signal feature values of the reference video signal from the feature value calculating part (12), retrieves correction information corresponding to the received video signal feature values from the correction information storing part (13), and transfers the retrieved correction information to a correcting part (15). According to the transferred correction information, the correcting part (15) corrects the subjective quality estimated by the subjective quality estimating part (11).