Reference-Free Video Quality Assessment Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer-implemented methods for evaluating video quality are limited as they require structural analysis of source video content compared to suboptimal content, necessitating similar frame rates and dimensions, and fail to account for subjective user perceptions.
Innovation Solution
A machine learning system, such as a convolutional neural network (CNN), is trained using user input quality scores and video characteristics to simulate subjective quality evaluation without requiring information about the correct appearance of the video, allowing for the determination of an estimated score based on video characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If structural analysis of source video content is performed to evaluate video quality, then measurement precision is improved, but device complexity and difficulty of detecting and measuring increase due to requirements for similar frame rates and dimensions
Solution Approach 1:
The patent uses a trained machine learning model (neural network) that has learned the characteristics of high-quality video content during training. During evaluation, the model takes the suboptimal video and compares it against the learned patterns from high-quality reference data stored in its weights, eliminating the need for actual reference video content and complex structural analysis.
Solution Approach 2:
The patent replaces traditional mechanical/algorithmic video comparison methods with a machine learning-based evaluation system. Instead of pixel-by-pixel structural analysis requiring matching frame rates and dimensions, the neural network processes video characteristics and directly outputs quality scores, simplifying the evaluation process.
2Measurement precision
If reference-based video quality evaluation is used, then measurement precision is improved, but loss of information increases due to requirement for source video content and reference material
Solution Approach 1:
The machine learning model encapsulates the knowledge of high-quality video characteristics within its trained parameters. During evaluation, only the suboptimal video needs to be provided, and the model internally retrieves the learned high-quality patterns from its weights, eliminating the need to transmit or store actual reference video content.
Solution Approach 2:
The patent extracts the essential quality evaluation capability from the reference video content itself and encapsulates it in the machine learning model. The reference information is taken out of the evaluation process and embedded in the model's training data, allowing evaluation without requiring the actual reference material during runtime.
3Ease of operation
If traditional video quality evaluation methods are used, then ease of operation is improved, but reliability worsens due to failure to account for subjective user perceptions
Solution Approach 1:
The machine learning model was trained using feedback from subjective user evaluations of video quality. During training, the model learned to predict human quality perceptions by analyzing the relationship between video characteristics and user scores, enabling it to produce reliability metrics that reflect actual user experience.
Solution Approach 2:
The patent transforms the evaluation approach by changing from objective structural parameters (frame rate, resolution matching) to subjective quality parameters captured through machine learning. The model processes video characteristics and outputs quality scores that mirror human perception, improving reliability while maintaining operational simplicity.
Data Source
AI summary
A machine learning system is trained to determine scores indicative of a quality of video data based on the characteristics of the video data, without requiring information regarding the correct appearance or other aspects of the video content. To train the machine learning system, users input scores for videos having predetermined quality scores, videos that have been previously seen by the users, and videos that have not been previously seen by the users. The differences between a user's score and a predetermined score or a score previously input by the user are used to determine a consistency metric. The scores and consistency metrics determined for a group of users, and the video characteristics of the videos presented to the users, are used to train the machine learning system to determine scores indicative of the quality of a video based on the characteristics of the video.