Image Super-Resolution Using Deep Convolutional Networks¶
Why this mattered¶
TBD
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.
Related¶
- cite → ImageNet: A large-scale hierarchical image database — SRCNN relies on ImageNet-era large-scale visual recognition progress as evidence that convolutional networks can learn powerful image representations.
- cite → Gradient-based learning applied to document recognition — SRCNN adapts the convolutional-network learning paradigm established by LeNet-style document recognition to low-level image super-resolution.
- cite → Image quality assessment: from error visibility to structural similarity — SRCNN evaluates super-resolved images using SSIM, the structural-similarity image-quality metric introduced by Wang and colleagues.
- cite → Backpropagation Applied to Handwritten Zip Code Recognition — SRCNN uses backpropagation to train its convolutional filters end to end, following the method demonstrated for handwritten ZIP-code recognition.
- cite → ImageNet classification with deep convolutional neural networks — SRCNN cites AlexNet-style deep convolutional image classification as motivation for applying deep CNNs to image super-resolution.
- cite ← Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network — SRGAN extends SRCNN's deep-learning approach to single-image super-resolution by replacing pixel-loss optimization with adversarial and perceptual losses.
- enables ← ImageNet: A large-scale hierarchical image database — ImageNet-scale natural-image data enabled SRCNN to train and evaluate deep convolutional super-resolution on realistic image distributions.
- enables ← Gradient-based learning applied to document recognition — LeNet's convolutional weight sharing and end-to-end gradient training provided the CNN template adapted by SRCNN for super-resolution.
- enables ← Image quality assessment: from error visibility to structural similarity — SSIM provided the perceptual image-quality metric used to evaluate reconstruction fidelity in SRCNN super-resolution experiments.
- enables ← Backpropagation Applied to Handwritten Zip Code Recognition — Backpropagation for convolutional handwritten-digit recognition demonstrated end-to-end learned image filters later reused in SRCNN-style vision networks.