Augmented Super Resolution GAN (ASRGAN) for Image Enhancement Through Reinforced Discriminator

Image super-resolution improves an image resolution and provides better detail. Super resolution achieved through Generative Adversarial Networks has been found to produce good resolution but has few challenges such as discrimination ability, stability and convergence of SRGAN. To address these challenges, we propose an Augmented Super resolution Generative Adversarial Network (ASRGAN) that incorporates reinforcement learning (RL) techniques. In ASRGAN, we introduce an RL agent into the discriminator component to enhance its discrimination capability and adapt its behavior to optimize the training process. The proposed ASRGAN signifies that the SRGAN has been expanded with reinforcement learning, which can have implications for the training stability, discrimination ability, convergence, and fine-grained control of the super resolution process. The generalization capabilities of ASRGAN across diverse image types, highlighting its adaptability has also been brought through this work.