Objective Video Quality Assessment Based on Machine Learning for Underwater Scientific Applications.

Area of Science:

• Oceanic research
• Network engineering
• Multimedia communications

Background:

• Underwater networks (UWNs) face severe bitrate limitations, impacting video service utility.
• Quality of Experience (QoE) for ocean scientists is sensitive to minor video parameter variations.
• Objective Video Quality Assessment (VQA) is crucial for UWN planning and real-time adaptation.

Purpose of the Study:

• To develop specialized objective VQA models for UWNs.
• To enhance the reliability and utility of video services in oceanic research.

Main Methods:

• Utilized machine learning techniques.
• Trained models with user data from subjective tests.
• Developed two distinct VQA models tailored for UWN constraints.

Main Results:

• Both models accurately estimate Mean Opinion Score (MOS) for video quality.
• The second model additionally provides a distribution function for user scores.
• Demonstrated successful quality estimation despite limited network capacity.

Conclusions:

• The developed VQA models are effective for UWN environments.
• These models support network planning and real-time quality adaptation for oceanic research.
• Improved VQA enhances the usability of video services in underwater research.