No-Reference Video Quality Assessment Using Natural Spatiotemporal Scene Statistics.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Spatiotemporal representations are crucial for video analysis tasks like quality assessment and action recognition.
• Existing methods often require reference videos, limiting practical applications.

Purpose of the Study:

• To develop a robust video representation for analyzing spatiotemporal statistics.
• To propose a no-reference video quality assessment (NRVQA) algorithm based on this representation.
• To evaluate the algorithm's performance on diverse video datasets.

Main Methods:

• Modeling spatiotemporal statistics of mean subtracted and contrast normalized (MSCN) coefficients using an asymmetric generalized Gaussian distribution (AGGD).
• Utilizing AGGD model parameters as features for distortion discrimination.
• Employing support vector regression (SVR) for the NRVQA task.

Main Results:

• The proposed AGGD model effectively captures video spatiotemporal statistics.
• AGGD parameters demonstrate strong discriminative power for video distortions.
• The SVR-based NRVQA algorithm achieves competitive performance on synthetic distortions and acceptable performance on authentic distortions.

Conclusions:

• The proposed parameterized statistical model offers a robust approach to video representation.
• The AGGD-based NRVQA method shows promise for practical video quality evaluation without reference videos.