An Optical Flow-Based Full Reference Video Quality Assessment Algorithm.

Area of Science:

• Computer Vision
• Image and Video Processing
• Perceptual Quality Assessment

Background:

• Assessing video quality is crucial for various applications, including streaming and broadcasting.
• Existing full-reference video quality assessment (FR-VQA) methods often rely on complex models or specific distortion types.
• Optical flow, representing pixel movement, contains information about video content and potential distortions.

Purpose of the Study:

• To develop a simple, effective, and parameter-free optical flow-based FR-VQA algorithm.
• To assess the perceptual quality of natural videos by analyzing deviations in optical flow statistics.
• To create a method competitive with state-of-the-art FR-VQA algorithms across diverse distortion types.

Main Methods:

• Characterized local optical flow statistics using mean, standard deviation, coefficient of variation (CV), and minimum eigenvalue (λ min).
• Estimated temporal distortion via changes in CV and correlation of λ min between reference and distorted flows.
• Utilized a multi-scale structural similarity index for spatial quality estimation and pooled results for a spatio-temporal score.

Main Results:

• The proposed flow similarity index algorithm demonstrated competitive performance against state-of-the-art methods on multiple benchmark databases (LIVE SD, EPFL Polimi SD, LIVE Mobile HD).
• The method effectively assessed distortions including compression, packet-loss, wireless errors, and rate-adaptation.
• The algorithm proved flexible, parameter-free, and independent of the specific optical flow computation method.

Conclusions:

• Optical flow statistics provide a robust basis for perceptual video quality assessment.
• The flow similarity index offers a simple yet effective FR-VQA solution adaptable to various distortions.
• Even coarser block motion vectors can yield acceptable FR-VQA results, indicating the robustness of the approach.