자연의 영상에서 문자 합성 모델을 사용한 문자 검출

Abstract

In this paper, we propose a pipeline to detect the text block data using the text synthesis model. We generate various types of texts using GAN. With the given generated text, we naturally synthesized natural images through a single image depth estimation and segmentation process. The generated synthesis image helps to secure the text block data that robustly adapts to the natural image. In our experiments, it is shown that the proposed algorithm demonstrates robust text detection performance