Image Restoration from Low Bit-Rate Quantization Based on Reinforcement Learning

The image restoration is an important issue in some application scenarios where only the limited storage or bandwidth are available. In this paper, we present a method using pixelwise reinforcement learning to restore the uniformly quantized images. The results are evaluated using various metrics, including SSIM, PSNR and HDR-VDP-3, and the performance comparison with other techniques is provided. The experimental results demonstrate the better visual effect and image quality of the proposed method. It also shows the potential to deal with more complicated high dynamic range imaging problems.