GaoZheng Liu's Dissertation Defense
PhD candidate GaoZheng Liu will present his dissertation defense, “Reading Order-Aware OCR for Visually Rich Historical Documents: Towards Better Accessibility and Analysis.” Liu's dissertation committee includes Professor J. Stephen Downie (Chair), Teaching Assistant Professor Jill Naiman, Professor Jiangping Chen, and Associate Professor Vetle Torvik.
Abstract
Optical Character Recognition (OCR) is a long-standing task in extracting text from images. While OCR systems achieve high accuracy in character and word recognition, reading order remains unresolved, particularly for visually rich documents. Reading order errors disrupt text sequence and degrade downstream processing. Although prior work has addressed aspects of reading order, systematic error and impact analysis, as well as dedicated evaluation resources, are still lacking. Existing modeling approaches also remain limited in handling the diverse and complex reading order patterns found in real documents. This thesis addresses these gaps through three contributions. First, a structural analysis of reading order errors is conducted to identify their patterns, causes, and effects. Second, a multi-granularity modeling framework is developed to represent reading order with greater flexibility. Third, new benchmark datasets and evaluation metrics are constructed to assess reading order performance in historical documents. Together, these contributions support more reliable reading order–aware OCR for improved accessibility and analysis of visually rich historical documents.
Questions? Contact GaoZheng Liu.