Super Resolution for QR Code Images

In this paper, we propose an image denoising and a super resolution method for converting unreadable low resolution QR code images into readable high resolution ones. We propose an image denoising and a super resolution method using a simple CNN-based model. Image denoising using CNN-based image generation models can be applied to various noises by training model with different levels of noise. On the other hand, denoising using the conventional image processing method can only be applied to the specific type of noise. Therefore, image denoising using CNN is considered to be superior to image processing methods in terms of generalization performance. We further implement super resolution method to convert into high resolution images. We propose QRCNN and QRGAN, simple image denoising and super resolution models for QR code images. QRCNN and QRGAN are based on SRResNet and SRGAN, respectively. However, they both have simpler structure like SRCNN. Given QR code images for the dataset, we can reduce the computational complexity and memory usage by modifying model structure.