Low-Resolution Image Enhancement using Generative Adversarial Networks

Low-Resolution Image Enhancement using Generative Adversarial Networks

Poster 209

Low-resolution image enhancement has long been in the public’s consciousness. Television shows, movies, and other forms of fiction have long imagined improving the quality of blurry or distorted images. This bold new technology has finally become available with the help of machine learning. Generative Adversarial Networks (GANs) hold the potential to achieve that which was thought to only exist in fiction. The GANs ability to enhance image quality without affecting the cost of the image gives it a unique opportunity to benefit many industries. Law enforcement could identify criminals with better accuracy. Internet Service Providers could transmit at lower resolutions, with the signal being upscaled to the desired resolution on the client’s side, reducing latency and bandwidth. Our research focuses on improving the performance of current GAN networks, such as Super Resolution GAN (SRGAN). Further optimization can be achieved by modifying the architecture and training configuration of the network. These improvements have a high likelihood to increase the speed and quality of the image, while decreasing the cost.