Two-Level Approach for No-Reference Consumer Video Quality Assessment.

Area of Science:

• Computer Vision
• Signal Processing
• Multimedia Systems

Background:

• Consumer devices enable real-time video capture and sharing, increasing demand for no-reference video quality assessment (NR-VQA).
• Existing NR-VQA models primarily address coding/transmission distortions, not capture impairments common in consumer videos.
• Current accurate NR-VQA methods are often computationally intensive, limiting practical applications.

Purpose of the Study:

• To develop a computationally efficient and accurate NR-VQA method for consumer-generated video content.
• To address the limitations of existing models in handling capture impairments and computational complexity.

Main Methods:

• A novel learning-based NR-VQA approach utilizing two levels of feature computation.
• Low-complexity features are extracted from the full video sequence.
• High-complexity features are extracted from a subset of representative frames identified using the low-complexity features.

Main Results:

• The proposed two-level feature method demonstrates superior accuracy in predicting subjective video quality compared to benchmark methods.
• The method was validated using three public video quality databases.
• Achieved comparable accuracy to the best prior method but with significantly reduced computational cost.

Conclusions:

• The proposed two-level feature approach offers an effective and efficient solution for NR-VQA of consumer videos.
• This method is practical for real-world applications due to its balance of accuracy and computational efficiency.
• Advances NR-VQA for the growing volume of user-generated video content.