Image Super-Resolution Using Deep Convolutional Networks

Abstract

We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low/high-resolution images. The mapping is represented as a deep convolutional neural network (CNN) that takes the low-resolution image as the input and outputs the high-resolution one. We further show that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network. But unlike traditional methods that handle each component separately, our method jointly optimizes all layers. Our deep CNN has a lightweight structure, yet demonstrates state-of-the-art restoration quality, and achieves fast speed for practical on-line usage. We explore different network structures and parameter settings to achieve trade-offs between performance and speed. Moreover, we extend our network to cope with three color channels simultaneously, and show better overall reconstruction quality.

14 pages, 14 figures, journal

References in corpus (3)

• Speeding up Convolutional Neural Networks with Low Rank Expansions
• DeepID-Net: multi-stage and deformable deep convolutional neural networks for object detection
• Convolutional Neural Networks at Constrained Time Cost

Cited by in corpus (40)

• Low Dose CT Image Denoising Using a Generative Adversarial Network with Wasserstein Distance and Perceptual Loss
• Perceptual Losses for Real-Time Style Transfer and Super-Resolution
• Unsupervised Cross-Domain Image Generation
• Systematic evaluation of CNN advances on the ImageNet
• Super Resolution Convolutional Neural Network Models for Enhancing Resolution of Rock Micro-CT Images
• Multi-Stage Progressive Image Restoration
• Deep Cascaded Bi-Network for Face Hallucination
• An Application of Generative Adversarial Networks for Super Resolution Medical Imaging
• Sparsity-Based Super Resolution for SEM Images
• Probabilistic Autoencoder
• One-Shot Free-View Neural Talking-Head Synthesis for Video Conferencing
• Global-Local Face Upsampling Network
• Uncertainty Quantification with Generative Models
• (ASNA) An Attention-based Siamese-Difference Neural Network with Surrogate Ranking Loss function for Perceptual Image Quality Assessment
• Frequency Domain Loss Function for Deep Exposure Correction of Dark Images
• The Pace of Artificial Intelligence Innovations: Speed, Talent, and Trial-and-Error
• An Iterative Convolutional Neural Network Algorithm Improves Electron Microscopy Image Segmentation
• Deep Joint Rain Detection and Removal from a Single Image
• Language Independent Single Document Image Super-Resolution using CNN for improved recognition
• Spatial-Temporal Super-Resolution of Satellite Imagery via Conditional Pixel Synthesis
• Sketch2code: Generating a website from a paper mockup
• A Compact DNN: Approaching GoogLeNet-Level Accuracy of Classification and Domain Adaptation
• Prediction-assistant Frame Super-Resolution for Video Streaming
• WiSoSuper: Benchmarking Super-Resolution Methods on Wind and Solar Data
• Global and Local Alignment Networks for Unpaired Image-to-Image Translation
• Arbitrary Scale Super-Resolution for Brain MRI Images
• Disentangle Perceptual Learning through Online Contrastive Learning
• Adaptive Loss Function for Super Resolution Neural Networks Using Convex Optimization Techniques
• Towards WARSHIP: Combining Components of Brain-Inspired Computing of RSH for Image Super Resolution
• TLGAN: document Text Localization using Generative Adversarial Nets
• Multilinear Map Layer: Prediction Regularization by Structural Constraint
• Separating Content and Style for Unsupervised Image-to-Image Translation
• Burst Image Restoration and Enhancement
• Memory-Efficient Hierarchical Neural Architecture Search for Image Restoration
• Theory of Generative Deep Learning : Probe Landscape of Empirical Error via Norm Based Capacity Control
• Finding Correspondences for Optical Flow and Disparity Estimations using a Sub-pixel Convolution-based Encoder-Decoder Network
• High Diversity Attribute Guided Face Generation with GANs
• Deep Residual Networks with a Fully Connected Recon-struction Layer for Single Image Super-Resolution
• Fine-scale Surface Normal Estimation using a Single NIR Image
• Compression Artifacts Reduction by a Deep Convolutional Network