Image Super-Resolution with Text Handling Via Generative Adversairal Network
생성적 적대 네트워크를 이용한 글자 특성이 고려된 초해상도 영상 복원

Jaepil Yu, Hyeongseok Son, Sunghyun Cho, Seungyong Lee
2018 KIISE Transactions on Computing Practices  
We propose a text-image super-resolution method using a generative adversarial network (GAN). Because previous super-resolution methods mainly learned properties of natural images, the quality of restored text regions in a super-resolution image is relatively low. The characteristics of text images and natural images are different, requiring an additional process to treat property of texts. We solve the text-image super-resolution problem by training natural images and text images separately.
more » ... mages separately. We add some texts to the dataset and use them for training the network to handle tex regions as well as natural-image regions. Experimental results show that the proposed network produces better text-image quality in the super-resolution results.
doi:10.5626/ktcp.2018.24.8.405 fatcat:lr5uqgsodnhjxk5srzyfpuebcm