Super-Resolution of Text Image Based on Conditional Generative Adversarial Network

To generate high-resolution text images from available low-resolution ones is of great value to many text-related applications, especially text recognition. In this paper, we propose an effective super-resolution method for text images based on Conditional Generative Adversarial Network (cGAN). Specifically, we improve the cGAN model by removing the Batch Normalization layers and introducing the Inception structure to make it more suited to the text image super-resolution task, which contribute to the overall enhanced performances of the proposed method relative to the original cGAN model. Experiment results on public dataset demonstrate the effectiveness of the proposed method.