Non-Reference Video Quality Evaluation Using CNN-RNN Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for evaluating video quality in a non-reference setting are inefficient and inaccurate, often resulting in large errors between predicted and perceived quality due to the complexity of designing algorithms without an original video, and face challenges in real-world applications where original videos are rarely available.
Innovation Solution
A method utilizing a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to evaluate video quality based on pixel values, with selective learning and pre-learning of convolutional layers to prevent over-fitting, and using error backpropagation for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If non-reference method is used to evaluate video quality without original video, then applicability in real communication environment is improved, but measurement precision deteriorates due to large error between predicted quality and actual perceived quality
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model using full-reference video quality data before deploying it for non-reference evaluation. The model learns from paired original and received video data during a training phase, enabling it to predict quality accurately even when only received video is available during actual deployment. This preliminary learning phase stores quality assessment knowledge that can be applied later without requiring original video.
Solution Approach 2:
The patent introduces an intermediary mechanism by using a neural network model as a bridge between full-reference training data and non-reference evaluation scenarios. The model acts as an intermediary that translates patterns learned from original-video-based training into quality predictions for scenarios without original video, effectively mediating between the two approaches and enabling accurate non-reference evaluation.
2Measurement precision
If full-reference method is used to evaluate video quality with original video, then measurement precision is improved with smaller error, but device complexity increases and applicability deteriorates since original video is rarely available
Solution Approach 1:
The patent uses preliminary action by training the neural network model in advance using full-reference video quality data. During this training phase, the model learns to associate video characteristics with quality metrics using both original and received videos. Once trained, the model can perform quality evaluation in real-time using only received video, eliminating the need for original video during actual deployment while maintaining high measurement precision.
3Device complexity
If simple algorithm with dozens of KPIs is used for non-reference evaluation, then device complexity is reduced, but measurement precision deteriorates due to large prediction error
Solution Approach 1:
The patent applies mechanics substitution by replacing the traditional mechanical algorithmic approach (using dozens of hand-crafted KPIs and mathematical algorithms) with a neural network-based system. Instead of manually designing and computing multiple quality indicators through complex algorithms, the system uses a trained neural network that automatically learns relevant features and patterns from data, substituting manual feature engineering with automated learning while improving both efficiency and accuracy.
Data Source
AI summary
A method for evaluating video quality based on a non-reference video to evaluate the quality of a received video using artificial intelligence composed of convolutional neural networks (CNN) and recurrent neural networks (RNN) capable of setting a learning range, includes: (a) extracting RGB values for one frame of a video; (b) obtaining an output by providing the extracted RGB values to CNN No. 1; (c) obtaining an output by providing the extracted RGB values n times to a CNN; (d) repeating (a) to (c) for all frames and merging outputs of all CNNs; (e) obtaining the output of the RNN with the time dimension reduced to 1 after passing the merged output value to the RNN; and (f) after applying a regression algorithm to the final output of the RNN, predicting this value as video quality value.