ST-GREED: Space-Time Generalized Entropic Differences for Frame Rate Dependent Video Quality Prediction

Area of Science:

• Computer Vision
• Signal Processing
• Perceptual Quality Assessment

Background:

• Video quality assessment (VQA) traditionally focuses on standard frame rates.
• High frame rate (HFR) videos present unique challenges for VQA due to temporal information density.
• Understanding the interplay between frame rate, compression, and perceived video quality is crucial.

Purpose of the Study:

• To develop an objective VQA model capable of assessing video quality across a wide range of frame rates, including HFR content.
• To investigate how frame rate and compression interact to influence perceived video quality.
• To propose a novel computational approach for frame rate-dependent VQA.

Main Methods:

• Devised Space-Time GeneRalized Entropic Difference (GREED), an objective VQA model.
• Analyzed statistics of spatial and temporal band-pass video coefficients.
• Modeled band-pass responses using generalized Gaussian distribution (GGD) and calculated entropy variations for quality assessment.

Main Results:

• GREED achieved state-of-the-art performance on the LIVE-YT-HFR Database compared to existing VQA models.
• The model demonstrated high generalizability, performing competitively on standard, non-HFR VQA databases.
• Entropy variations under the GGD model effectively captured quality degradation from frame rate changes.

Conclusions:

• GREED provides a robust and accurate method for frame rate-dependent VQA, outperforming current models on HFR content.
• The model's features are generalizable, offering broad applicability across different video types.
• The findings advance the understanding of perceptual quality in HFR video and provide a valuable tool for objective quality evaluation.