document-figures

To extract figures from a PDF with provenance, use a tool that pulls embedded images and renders page visuals while preserving source reference information. This process outputs a Figure Ledger linking each extracted figure to its exact page or slide location for auditability.

A Figure Ledger is a manifest output that records extracted figures with verifiable provenance, including IDs, source references, and recovered captions. It helps document summarization by allowing downstream workflows to cite each specific figure as evidence via linked Claim Ledger IDs.

Yes, you can automatically generate Mermaid diagrams when suitable figures are missing. This approach uses deterministic Mermaid generation guided by an information test to create structural diagrams for sequential, decision, state, and entity-relationship claims rather than decorated bullets.

Yes, figure extraction works with Office documents and web pages. It applies to converting PDFs, DOCX, PPTX, and web pages into Figure Ledgers, pulling embedded images and rendering visuals while preserving source reference information across these formats.

To extract images and recover captions from DOCX or PPTX files, you need dependencies like puppeteer, poppler-utils, xmllint, libxml2, and unzip. These tools support rendering visuals, parsing XML structures, and decompressing office document archives.

Recovered figure captions are flagged or reconstructed during extraction to maintain integrity and auditability. The tool explicitly flags reconstructed captions when original alt-text is missing, ensuring downstream summarization workflows know the confidence level of the recovered text.