最近在研究文字图像的超分辨率,下载了几篇论文,准备略读一下,每篇文章写个笔记做记录。
主要贡献:
A non-training text image super-resolution generating network TextSRDIP is proposed. By using the DIP network model, a new loss function is designed to highlight the high- frequency details of the image, aiming at paying more attention to the edge details of the image in the super- resolution reconstruction of the text image.After testing, the average reconstruction PSNR of the network for text image super resolution is 27.95db, and the optical character recognition accuracy is 73.3%, which is 4.45db and 12.66% higher than the traditional bicubic interpolation algorithm. Compared with the learning - b