. Super Resolution Reading List

Super Resolution Reading List

Early Learning-based Methods

  • Learning Low-Level Vision.
  • Example-based super-resolution.
  • Super-resolution through neighbor embedding.

Sparsity-based Methods

  • A Statistical Prediction Model Based on Sparse Representations for Single Image Super-Resolution
  • On Single Image Scale-Up Using Sparse-Representations
  • Image Deblurring and Super-resolution by Adaptive Sparse Domain Selection and Adaptive Regularization
  • Image Super-Resolution Via Sparse Representation

Anchored Neighborhood Regression Method

  • ANR: Anchored Neighborhood Regression for Fast Example-Based Super-Resolution
  • A+: Adjusted Anchored Neighborhood Regression for Fast Super-Resolution
  • ANR 2.0: Seven ways to improve example-based single image super resolution
  • ARN: Anchored Regression Networks applied to Age Estimation and Super Resolution

Deep Learning Method

Based on CNN

  • SRCNN: Image Super-Resolution Using Deep Convolutional Networks.
    This is the first paper which use CNN to handle SR problem.
  • VDSR: Accurate Image Super-Resolution Using Very Deep Convolutional Networks.
    This paper use residual learning and high learning rate to increase CNN layers up to 20 layers.
  • ESPCN: Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network.
    This paper remove bicubic interpolation operator by a smart-designed sub-pixel convolution layer. The feature map are extracted in the LR space. The designed sub-pixel convolution layer is very beautiful and the details can be found in supplemental material.
  • LapSRN: Fast and Accurate Image Super-Resolution with Deep Laplacian Pyramid Networks.
  • EDSR: Enhanced Deep Residual Networks for Single Image Super-Resolution.
    This is the best result in 2017. However, this paper did not bring up new method to handle SR problem.
  • SRDenseNet: Image Super-Resolution Using Dense Skip Connections. (2017 ICCV)
    This paper use DenseNet blocks, skip connect and deconvolution in the model to achieve new state of the art.

Based on RNN

  • DRCN: Deeply-Recursive Convolutional Network for Image Super-Resolution.
    This paper use Recursive Convolutional Network to handle SR problem.

Based on GANs

  • SRGAN: Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network.
    This is the first paper which use GAN to handle SR problem.
  • EnhanceNet: Single Image Super-Resolution Through Automated Texture Synthesis.
    This paper investigate different models (CNN & GAN) and different loss functions in SR problem. It bring up a question whether GAN is useful in SR problem.
  • MCGAN: Learning to Super-resolve Blurry Face and Text Images.
    This paper join super-resolution and deblurring into a single GAN network.

Based on Autoregressive Models

  • Pixel CNN: Pixel Recursive Super Resolution. (2017)
    This paper based on Pixel-CNN model which I have listed in "Generative Image Modeling Reading List".