Single image Super Resolution with Infrared Imagery and Multi Step Reinforcement Learning
Technical Report,24 Jul 2017,25 Aug 2020
University of Dayton Dayton United States
Pagination or Media Count:
Recent studies have shown that Deep Learning DL algorithms can significantly improve Super Resolution SR performance.Single image SR is useful in producing High Resolution HR images from their Low Resolution LR counterparts. The motivationfor SR is the potential to assist algorithms such as object detection, localization, and classification. Insufficient work has beenconducted using Generative Adversarial Networks GANs for SR on infrared IR images despite its promising ability to increaseobject detection accuracy by extracting more precise features from a given image. This work adopts the idea of a relativistic GANthat utilizes Residual in Residual Dense blocks RRDBs for feature extraction, a novel residual image addition, and a PixelTransposed Convolutional Layer PixelTCL for up-sampling. Recent work has validated the use of GANs for Visible Light VLimages, making them a strong candidate. The inclusion of these components produce more realistic and natural features while alsoreceiving superior metric values. Supplemental research applies a multi-agent Reinforcement Learning RL algorithm to SingleImage Super-Resolution SISR, creating an advanced ensemble approach for combining powerful GANs. In our implementationeach agent chooses a particular action from a fixed action set comprised of results from existing GAN SISR algorithms to update itspixel values. The pixel-wise arrangement of agents and rewards encourages the algorithm to learn a strategy to increase theresolution of an image by choosing the best pixel values from each option.