WCDANN: A Lightweight CNN Post-Processing Filter for VVC-Based Video Compression

In this paper, we propose a weakly connected dense attention neural network for compression artifact removal, called WCDANN. WCDANN is a convolutional neural network (CNN)-based post-processing filter to enhance the quality of versatile video coding (VVC)-decoded videos without requiring any codec c...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2023-01, Vol.11, p.1-1
Main Authors: Zhang, Hao, Jung, Cheolkon, Zou, Dan, Li, Ming
Format: Article
Language:English
Subjects:
Artificial neural networks
attention
Codec
convolutinal neural network
Convolution
convolutional neural network
Convolutional neural networks
depthwise separable convolution
Feature extraction
Image coding
in-loop filter
Lightweight
post-processing
Transform coding
Video coding
Video compression
Video post-production
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this paper, we propose a weakly connected dense attention neural network for compression artifact removal, called WCDANN. WCDANN is a convolutional neural network (CNN)-based post-processing filter to enhance the quality of versatile video coding (VVC)-decoded videos without requiring any codec changes. WCDANN consists of several weakly connected dense attention blocks (WCDABs) based on residual learning, which takes the compressed video after codecs as the input. We use depthwise separable convolution for WCDANN as the basic convolution unit to generate a lightweight model. Moreover, we introduce attention mechanisms into the proposed filter to capture important features. Experimental results show that WCDANN achieves good performance in Bjøntegaard Delta Bit Rate (BD-BR). Compared with VTM-11.0-NNVC anchor, WCDANN achieves average 2.81%, 4.12% and 3.81% BD-rate reductions for Y channel on A1, A2, B, C, D and E classes in RA, AI and LDP configurations, respectively.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2023.3301145