Hierarchical Layout Model and Reading Order Detection in Multi-article Non-Manhattan Page

Focusing on document image understanding(DIU) and text typesetting for newspaper, a hierarchical layout model(HLM) with a new page object called PMRegion is defined, and spatial relations are established in each layout layer for non-Manhattan page with multi-article. Based on the model, a quick algorithm for page top-down decomposition and unambiguous reading order detection is presented. They are proven to be effective in a specialized typesetting system and also helpful to go deep into DIU.