We wish to analyze scans of documents with text (non-handwritten) and images with very broad range of arrangements/structures in different languages. The first problem we try to solve, is extracting text and identifying and separating titles, subtitles and text bodies.
At the moment we are doing a literature research. There is plenty of literature about deep learning, computer vision, optical character recognition or natural language processing but none of these are actually focused on optical recognition of the structure of text.
We wonder, what is the name of the discipline/field that deals with optical recognition of structure of text?
What are the state-of-the-art approaches and tools for solving these problems?
Optical Layout Recognition (OLR). A good example of an open-source tool for Layout Analysis and Region Extraction can be found here.