Generative Adversarial Network for Photo-Realistic single image super-resolution

Regardless the breakthrough in accuracy and speed of single image super-resolution using faster and deeper convolution neural networks, one central problem remains unsolved: how do we recover the finer texture details when we super-resolve at large up-scaling factors? Recent work has largely focused

Project Title

Generative Adversarial Network for Photo-Realistic single image super-resolution

Project Area of Specialization

Artificial Intelligence

Project Summary

Regardless the breakthrough in accuracy and speed of single image super-resolution using faster and deeper convolution neural networks, one central problem remains unsolved: how do we recover the finer texture details when we super-resolve at large up-scaling factors? Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios (PSNR) and failed to work at high frequencies. So, we present SRGAN, a Generative Adversarial network (GAN) for super-resolution (SR). The highly challenging task of estimating a high-resolution (HR) image from its low-resolution counterpart is referred as super-resolution (SR). To our knowledge, this is the first framework capable of inferring photo-realistic natural image for 4x up-scaling factors.

Project Objectives

•Estimating high-resolution (HR) image from its low-resolution (LR) counterpart image.

•Recover finer texture details at higher up-scaling factors.

•Cost Effective.

•Reduce human Effort.

Project Implementation Method

We present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4× up-scaling factors.We will be using generative adversarial networks as the main frame, including generator network and discriminator network. LR image is the generator network’s input, then the convolutional layers are responsible for extracting features. Subsequently, the feature map inputs residual model for non-linear mapping. Then the image is reconstructed through the up-sampling layer and convolutional layer. Next, the network outputs the reconstruction result. Finally, we input the fake and real HR images into discriminator network separately, which is responsible for discriminating the authenticity of image.

Benefits of the Project

• Cost Effective: 

As most of the work will be performed using deep learning and image processing and no need of high end capturing devices to attain high-resolution images.

• Reduce human Effort

Most of the work will be performed by deep learning and the machine which will be using the input LR images and producing HR images.

• Improve Efficiency

The system will improve efficiency due to the use of GANs. As these are the most efficient networks in deep learning algorithm.

Super Resolution received substantial attention from within the computer vision research community and has a wide range of applications.

Technical Details of Final Deliverable

Generative Adversarial Network for Photo-Realistic single image super-resolution uses 2 networks: fist one is generator and second one is adversarial netowrk including a discriminator network.. Generator produces some data based on the probability distribution and discriminator tries to guess weather data coming from input dataset or generator. Generator than tries to optimize the generated data so that it can fool the discriminator. The Final Deliverable is the Super Resolved High Quality image construscted form its low-resolution counterpart.

Final Deliverable of the Project

Software System

Core Industry

Security

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure, Peace and Justice Strong Institutions

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com