DRGNet: Enhanced VVC Reconstructed Frames Using Dual-Path Residual Gating for High-Resolution Video.

Area of Science:

• Computer Vision
• Image Processing
• Video Coding

Background:

• High-resolution video and internet traffic growth necessitate efficient video coding like H.266 (VVC).
• Video compression introduces artifacts and detail loss, degrading visual quality and user experience.
• Existing methods struggle with artifact reduction and detail preservation in high-resolution VVC streams.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) based post-processing method for enhancing VVC reconstructed frame quality.
• To effectively reduce compression artifacts, remove noise, and minimize detail loss in high-resolution videos.
• To improve the visual experience for users by delivering clearer and more detailed video content.

Main Methods:

• A high-resolution dual-path residual gating system utilizing deep feature extraction and fusion.
• Integration of convolutional blocks with gating mechanisms and residual connections.
• Selective feature preservation and artifact removal through combined gating and residual operations.

Main Results:

• Significant improvement in Peak Signal-to-Noise Ratio (PSNR) under identical bitrate conditions.
• Demonstrated BD-Rate improvements of 6.1% (RA), 7.36% (LDB), and 7.1% (AI) for the luma component.
• Enhanced visual quality with reduced artifacts and preserved details in VVC reconstructed frames.

Conclusions:

• The proposed CNN-based post-processing method effectively enhances VVC video quality.
• The dual-path residual gating system offers superior artifact reduction and detail preservation.
• The method provides a clearer, more detailed visual experience, optimizing user satisfaction.