Document Image Understanding

Document image understanding studies how to interpret scanned, photographed, and digital pages, from recognizing text to recovering layout, tables, diagrams, and meaning across pages. Common approaches include OCR paired with layout detection and reading-order models, specialized table or handwriting recognizers, and end-to-end vision-language models; retrieval-augmented systems use these outputs to answer questions with document evidence. Benchmarks test tasks such as text extraction, document question answering, table parsing, and multilingual or multi-page understanding. Models have improved on clean pages, but real-world degradation, rare scripts, complex layouts, and long-range evidence integration remain difficult, and strong text accuracy does not always translate into reliable downstream answers.