Machine Learning based Post Processing Artifact Reduction in HEVC Intra Coding

TL;DR

This paper introduces a deep learning-based post-processing method for HEVC intra coding that enhances artifact reduction and reduces bit rate, outperforming existing CNN approaches in quality metrics and efficiency.

Contribution

It proposes a novel variable filter size deep CNN architecture for SAO filtering, improving artifact removal and reducing overhead in HEVC intra coding.

Findings

Outperforms other CNN-based methods in PSNR and SSIM.

Achieves an average of 4.1% bit rate reduction over HEVC baseline.

Effective training with data augmentation and transfer learning.

Abstract

The lossy compression techniques produce various artifacts like blurring, distortion at block bounders, ringing and contouring effects on outputs especially at low bit rates. To reduce those compression artifacts various Convolutional Neural Network (CNN) based post processing techniques have been experimented over recent years. The latest video coding standard HEVC adopts two post processing filtering operations namely de-blocking filter (DBF) followed by sample adaptive offset (SAO). These operations consumes extra signaling bit and becomes an overhead to network. In this paper we proposed a new Deep learning based algorithm on SAO filtering operation. We designed a variable filter size Sub-layered Deeper CNN (SDCNN) architecture to improve filtering operation and incorporated large stride convolutional, deconvolution layers for further speed up. We also demonstrated that deeper architecture model can effectively be trained with the features learnt in a shallow network using data augmentation and transfer learning based techniques. Experimental results shows that our proposed network outperforms other networks in terms on PSNR and SSIM measurements on widely available benchmark video sequences and also perform an average of 4.1 % bit rate reduction as compared to HEVC baseline.