AUTOMATIC IMAGE CONTRAST ENHANCEMENT BASED ON REINFROCEMENT LEARNING
Abstract: Image contrast enhancement is a subjective problem depending on personal preference and subject field property. Every person has different perception on the assessment of an enhanced image quality. Thus, it is difficult to have one ideal outcome that satisfies every person with the existing conventional image enhancement techniques. In this paper, we proposed a simple and efficient reinforcement learning based image contrast enhancement method for personal preference. Our method consists of state, action, reward or punishment definition, and policy learning. We have implemented Q-learning and State Action Reward State Action (SARSA) algorithms. The training process is easy for any user by clicking some buttons in our developed graphical user interface (GUI). The experimental results demonstrate good performance of our proposed method in this paper.
First algorithm on the training
Second algorithm on the training
training evaluation
Test result
