Multimedia Session MOS Estimator Using ML Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating Mean Opinion Score (MOS) in multimedia sessions, particularly over HTTP delivery, are inadequate for complex delivery scenarios and do not effectively account for factors like rebuffering and adaptive bitrate, resulting in rudimentary quality assessment.
Innovation Solution
A method using machine learning techniques, specifically a random forest model, to predict multimedia session MOS by generating features from per time unit scores of video and audio quality, as well as buffering events, such as total buffering time and number of rebuffering events, to estimate the subjective quality of multimedia sessions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional MOS estimation models are used for HTTP delivery, then the assessment is simpler to implement, but the quality assessment is rudimentary and insufficient for complex delivery scenarios
Solution Approach 1:
The patent segments the quality assessment into multiple independent feature extraction components: video quality features, audio quality features, and buffering features. Each component processes specific aspects of the multimedia stream separately, then combines them for comprehensive MOS prediction. This segmentation allows the system to handle complex delivery scenarios while maintaining modularity and manageability.
Solution Approach 2:
The patent introduces a new dimension of analysis by incorporating buffering events (initial buffering and rebuffering) as a separate feature category alongside traditional video and audio quality metrics. This dimensional expansion enables the model to capture network delivery characteristics specific to HTTP streaming, transforming the assessment from purely media-quality focused to include delivery-performance dimensions.
2Productivity
If simple parameters are used to estimate video stream quality, then the estimation is faster and less computationally complex, but the accuracy is insufficient for complex delivery scenarios
Solution Approach 1:
The patent performs preliminary action by extracting and storing multiple pre-computed features from the multimedia stream before final MOS prediction. Video quality features (sharpness, noise, blockiness), audio quality features, and buffering features are all extracted and prepared in advance, allowing the random forest model to make accurate predictions without real-time computational overhead during actual quality assessment.
Solution Approach 2:
The patent changes parameters by transitioning from simple quality metrics to a multi-parameter feature set that includes detailed video quality indicators (sharpness, noise, blockiness), audio quality indicators, and buffering event characteristics. This parameter expansion enables accurate quality estimation while the random forest model efficiently processes these multiple parameters through trained decision trees.
3Measurement precision
If pixel-based metrics from decoded video stream are used, then the video quality assessment is more accurate, but the method requires full reference information from both sender and receiver ends
Solution Approach 1:
The patent extracts only the necessary quality indicators from the decoded video stream rather than requiring complete pixel-based comparison data. Specifically, it extracts key features such as sharpness, noise, and blockiness metrics from the received video, eliminating the need for sender-end reference information while maintaining accurate video quality assessment capability.
Data Source
AI summary
It is provided a method, performed by a MOS, Mean Opinion Score, estimator, for predicting a multimedia session MOS. The multimedia comprises a video and an audio, wherein video quality is represented by a list of per time unit scores of a video quality, an initial buffering event and rebuffering events in the video, and wherein audio quality is represented by a list of per time unit scores of audio quality. The method comprises: generating video features from the list of per time unit scores of the video quality; generating audio features from the list of per time unit scores of the audio quality; generating buffering features from the initial buffering event and rebuffering events in the video; and estimating a multimedia session MOS from the generated video features, generated audio features and generated buffering features by using machine learning technique.