λ-Domain Rate Control via Wavelet-Based Residual Neural Network for VVC HDR Intra Coding

Area of Science:

• Video compression
• Image processing
• Machine learning

Background:

• High Dynamic Range (HDR) video offers enhanced realism but poses compression challenges.
• Existing Versatile Video Coding (VVC) rate control algorithms are optimized for Standard Dynamic Range (SDR) and perform poorly on HDR content.
• Effective rate control is crucial for efficient HDR video delivery.

Purpose of the Study:

• To develop a data-driven rate control algorithm for VVC HDR intra frames.
• To address the limitations of current VVC rate control for HDR video.
• To improve the coding efficiency and visual quality of HDR video compression.

Main Methods:

• Analysis of HDR intra coding characteristics.
• Development of a piecewise Rate-Lagrange parameter (R-λ) model for HDR intra frames.
• Implementation of a wavelet-based residual neural network (WRNN) for CTU-level bit allocation optimization.
• Creation of a large-scale HDR dataset for training the WRNN.

Main Results:

• The proposed piecewise R-λ model accurately captures the rate-distortion relationship for HDR intra frames.
• The WRNN effectively predicts R-λ model parameters for optimized bit allocation.
• Experimental results demonstrate superior coding performance compared to state-of-the-art methods.
• The developed algorithm achieves better compression efficiency for VVC HDR intra coding.

Conclusions:

• The proposed data-driven λ-domain rate control algorithm significantly enhances VVC HDR intra frame coding.
• The WRNN-based approach offers a promising direction for deep learning applications in HDR video compression.
• The algorithm provides a robust solution for optimizing HDR video delivery.