Full Document Extraction
Current document capability
Current capability
A document model, not just a text dump. Full Document Extraction combines recognized text, page structure, and visual regions into ordered, typed information that can move into systems, review, and reconstruction.
- Prepare
Convert and normalize the source
- Map
Detect regions and visual classes
- Read
Recognize text and structural observations
- Structure
Assemble one typed document model
- 17 page-layout classes
- 26 figure subtypes
- Structured document output
More than recognized text
Deep document understanding
The result keeps the relationships that make the page useful rather than flattening everything into one stream of characters.
- Document hierarchy, language, direction, reading order, and outline
- Tables and cells when recognized, including spans and header semantics
- Recognized field-value pairs, checkbox states with labels when available, and decoded barcodes when returned
- Word, line, element geometry, confidence, and source provenance
The page, classified
17 page-layout classes
Nasaas distinguishes the parts of a page before assembling them into the final document model.
- Titles, section headings, and body text
- List items, captions, footnotes, formulas, and code
- Page headers, page footers, and document indexes
- Tables, forms, and key-value regions
- Selected and unselected checkbox regions
- Pictures and other visual regions
Visual content is part of the document
26 figure subtypes
Pictures can be classified by what they contain instead of being reduced to an anonymous image box.
- Marks: stamps, logos, signatures, and icons
- Images: photographs, screenshots, page thumbnails, and full-page images
- Charts: line, bar, pie, scatter, and box plots
- Technical visuals: flowcharts, engineering drawings, and chemical structures
- Maps and structured visuals: geographic and topographic maps, calendars, music, crossword puzzles, and visual tables
- Codes and remaining types: QR codes, barcodes, and other figures
Fast Structure-aware OCR
Available extraction mode
Use the fast mode when you want the same structured document model and page intelligence at a flat one credit per page, without the deeper layout rerun used to reconstruct table cells, field pairs, and label-bound checkbox states.
- Retains recognized text, words, lines, coordinates, page regions, and visual classifications
- Produces the same canonical structured document model and available exports
- Detects table, checkbox, form, and key-value regions without running the deeper layout reread
- Does not reconstruct table matrices, cells, or spans
- Does not pair field values or associate checkbox states with labels
- Costs a flat 1 credit per input page; default Full Extraction costs 1 credit on ordinary pages and 3 on pages that trigger the layout rerun