Image Resolution Enhancement using Multi-Step Reinforcement Learning, machine & deep leaning
This work tackles the problem of image resolution enhancement using reinforcement learning with pixel-wise rewards. After the introduction of the deep Q-network, deep reinforcement learning has been achieving great success. However, the applications of deep reinforcement learning for image processing are still limited. Therefore, I try to address this issue by implementing a multi-agent reinforcement learning network which in theory could outperform traditional image upsampling algorithms.
This work is largely inspired by a similarly innovative solution called pixelRL. Similarly to the aforementioned paper I treat each pixel as a separate agent, and the agent changes the pixel value by taking an action.
Proposed method on BSD68 dataset and evaluate the prediction results against bicubic interpolation. My initial results demonstrate that the proposed method achieves comparable or better performance, using PSNR quality measurement.