Adaptive Chroma Prediction Based on Luma Difference for H.266/VVC

Area of Science:

• Video Compression Technology
• Digital Signal Processing
• Computer Vision

Background:

• Cross-component chroma prediction is crucial for efficient video coding in H.266/VVC.
• Existing methods may not fully leverage the relationship between luma and chroma components.

Purpose of the Study:

• To develop an attention-based model for chroma prediction using luma differences.
• To enhance coding efficiency in H.266/VVC through adaptive chroma weighting.

Main Methods:

• Designed a luma difference-based chroma prediction (LDCP) model utilizing an attention mechanism.
• Employed luma differences (LDs) as input to a softmax function for nonlinear mapping to chroma weights.
• Implemented adaptive weighting through template-based (T-LDCP) and offline learning (L-LDCP) approaches.

Main Results:

• T-LDCP achieved BD-rate reductions of 0.34% (Y), 2.02% (Cb), and 2.34% (Cr).
• L-LDCP resulted in BD-rate savings of 0.32% (Y), 2.06% (Cb), and 2.21% (Cr).
• L-LDCP introduced minor encoding (2%) and decoding (1%) time increases with minimal hardware-friendly parallelization benefits.

Conclusions:

• LDCP effectively improves chroma prediction accuracy and coding efficiency in H.266/VVC.
• Both T-LDCP and L-LDCP offer significant bitrate savings with L-LDCP providing slightly better compression.
• The LDCP method is hardware-friendly due to its pixel-level parallelization capabilities.