Constructing Per-Shot Bitrate Ladders Using Visual Information Fidelity

Area of Science:

• Computer Science
• Signal Processing
• Multimedia Engineering

Background:

• HTTP Adaptive Streaming (HAS) enables dynamic video quality adjustments.
• Per-shot encoding optimizes video parameters scene by scene.
• Existing methods may not fully leverage content characteristics for optimal streaming.

Purpose of the Study:

• To develop a perceptually optimized method for constructing per-shot bitrate and quality ladders.
• To predict optimal video encoding parameters without compression or quality estimation.
• To improve video streaming efficiency and user experience.

Main Methods:

• Utilized an ensemble of low-level and Visual Information Fidelity (VIF) features.
• Developed a predictive model for per-shot bitrate and quality ladder construction.
• Compared the proposed method against fixed ladders, content-adaptive methods, and exhaustive encoding references.

Main Results:

• The proposed method achieved significant bitrate and quality improvements over fixed ladders.
• Performance showed minimal losses compared to exhaustive encoding reference ladders.
• Demonstrated substantial computational advantages during inference.

Conclusions:

• The developed method offers an efficient approach to perceptual optimization in video streaming.
• It enables high-quality video delivery at reduced bitrates by predicting optimal encoding ladders.
• This technique holds promise for enhancing bandwidth consumption and user streaming experiences.