A Discrete-Mapping-Based Cross-Component Prediction Paradigm for Screen Content Coding.

Area of Science:

• Video Coding Technologies
• Digital Signal Processing
• Computer Vision

Background:

• Modern video codecs utilize cross-component prediction to enhance intra-prediction efficiency.
• Existing methods, like linear and multi-model linear approaches, are optimized for camera-captured content.
• Screen content videos possess distinct characteristics necessitating specialized prediction models.

Purpose of the Study:

• To develop a novel cross-component prediction model specifically for screen content coding.
• To leverage the unique signal properties of screen content, where luma often uniquely determines chroma.
• To improve prediction accuracy and coding efficiency for screen content.

Main Methods:

• A discrete-mapping based cross-component prediction model is proposed.
• The model learns a discrete mapping from reconstructed luma-chroma pairs.
• A multi-filter approach is used to derive co-located luma values for enhanced accuracy.

Main Results:

• The proposed model achieves 2.61% Y, 3.51% U, and 3.92% V bit-rate savings.
• Savings are demonstrated over the Enhanced Compression Model (ECM) 4.0.
• The method shows effectiveness for text and graphics media under all-intra configuration with negligible complexity.

Conclusions:

• The discrete-mapping approach offers a tailored and effective solution for screen content prediction.
• This method significantly reduces bit-rate requirements for screen content video coding.
• The proposed model represents a pioneering advancement in specialized video compression techniques.