Underwater Color Correction via Deep Reinforcement Learning

Abstract

Due to light is scattered and absorbed while it traveling in water, underwater captured images often suffer from color cast and low visibility. To address the above problems, we propose a deep reinforcement learning (DRL) based method for underwater color correction, which explicitly indicates the step-by-step nature of the color correction process. We cast a single underwater image processing process as a Markov Decision Process with several existing simple traditional image processing operations and white balance methods as actions. In addition, the deep Q-learning network is used to establish our agent for underwater color correction by learning the optimal sequence of the actions. Through extensive experiments, we show that our method produces decent underwater color correction results and makes underwater image processing interpretable and extensible.