A Deep Learning-Based System for Document Layout Analysis
Hong Tai Tran, Nam Nguyen, Tuan Anh Tran, Xuan Toan Mai, Quoc Thang Nguyen · 2022
Document image understanding is an essential process in the digital transformation era. Those systems automatically convert a paper document to a digital document for storing and information extracting. In practice, document layout analysis is a critical step for the success of document image modeling. This paper introduces a page segmentation system based on deep neural networks. Our system uses two auto encoder-decoder networks to segment the text-line and non-text components simultaneously. The paragraph segmentation is then realized based on the text line and separator mask. Besides, the non-text elements are also identified. Our algorithm has been tested RDCL2019. Experimental results show that our method is more stable and more comfortable to adapt with a new format layout than the previous commercial and publishing systems.