Compression Artifacts Removal Using Convolutional Neural Networks

Abstract

This paper shows that it is possible to train large and deep convolutional neural networks (CNN) for JPEG compression artifacts reduction, and that such networks can provide significantly better reconstruction quality compared to previously used smaller networks as well as to any other state-of-the-art methods. We were able to train networks with 8 layers in a single step and in relatively short time by combining residual learning, skip architecture, and symmetric weight initialization. We provide further insights into convolution networks for JPEG artifact reduction by evaluating three different objectives, generalization with respect to training dataset size, and generalization with respect to JPEG quality level.

arXiv commentTo be published in WSCG 2016

References in corpus (4)

• Deep Residual Learning for Image Recognition
• DeepPose: Human Pose Estimation via Deep Neural Networks
• Learning a Convolutional Neural Network for Non-uniform Motion Blur Removal
• Accurate Image Super-Resolution Using Very Deep Convolutional Networks

Cited by in corpus (18)

• HDR image reconstruction from a single exposure using deep CNNs
• DR2-Net: Deep Residual Reconstruction Network for Image Compressive Sensing
• Implicit Dual-domain Convolutional Network for Robust Color Image Compression Artifact Reduction
• Video Frame Interpolation via Adaptive Separable Convolution
• One-Two-One Networks for Compression Artifacts Reduction in Remote Sensing
• Video Frame Interpolation via Adaptive Convolution
• S-Net: A Scalable Convolutional Neural Network for JPEG Compression Artifact Reduction
• BlockCNN: A Deep Network for Artifact Removal and Image Compression
• A New Convolutional Network-in-Network Structure and Its Applications in Skin Detection, Semantic Segmentation, and Artifact Reduction
• Image Reconstruction with Predictive Filter Flow
• Non-Local ConvLSTM for Video Compression Artifact Reduction
• No Multiplication? No Floating Point? No Problem! Training Networks for Efficient Inference
• Ultra High Fidelity Image Compression with ${\ell }_{\infty }$-constrained Encoding and Deep Decoding
• Boosting High-Level Vision with Joint Compression Artifacts Reduction and Super-Resolution
• Near-lossless ${\ell }_{\infty }$-constrained Image Decompression via Deep Neural Network
• The differential geometry of perceptual similarity
• Likelihood Landscapes: A Unifying Principle Behind Many Adversarial Defenses
• Machine Vision for Improved Human-Robot Cooperation in Adverse Underwater Conditions