Extract Tables from PDF

PDF documents store text as individually positioned glyphs rather than structured data like HTML tables or spreadsheet cells. A PDF file contains instructions like "draw the character 'A' at position (72, 540)" — there is no concept of rows, columns, or cell boundaries in the file format itself. This fundamental design characteristic makes table extraction from PDFs a coordinate geometry problem rather than a simple parsing task.

This tool implements the stream algorithm, one of two primary approaches to PDF table extraction. The stream algorithm works by analyzing text position coordinates without relying on visible grid lines or borders. It operates in three stages:

1. Row clustering — Text items are grouped by their Y-coordinate proximity. Items within 3 PDF points of the same vertical position are merged into a single logical row. Rows are then sorted from top to bottom (PDF uses a bottom-left coordinate origin, so higher Y values appear at the top of the page).
2. Column detection — Across all identified rows, the algorithm collects every text item's X starting position and clusters them into vertical alignment groups. Gaps of 8 or more points between X-position clusters indicate column boundaries. This produces a set of representative X positions defining the table's column structure.
3. Cell assignment — Each text item in every row is assigned to the nearest detected column based on X-coordinate proximity. When multiple text fragments map to the same cell, they are concatenated with spaces. Empty cells are preserved to maintain the rectangular grid structure.

The alternative approach, the lattice algorithm, detects visible grid lines by extracting vector drawing commands (lines and rectangles) from the PDF's content stream. While lattice detection is more accurate for bordered tables, it fails entirely on borderless tables — which account for a significant portion of tables in financial reports, academic papers, and government documents. The stream algorithm handles both bordered and borderless tables by relying solely on text alignment patterns.

CSV output follows RFC 4180 formatting conventions: fields containing commas, double quotes, or newlines are enclosed in double quotes, and embedded double quotes are escaped by doubling them. This ensures clean import into Microsoft Excel, Google Sheets, LibreOffice Calc, and programmatic CSV parsers.

Extracting tables from PDFs addresses a widespread data accessibility challenge across multiple industries:

• Financial analysis — Annual reports, earnings statements, and SEC filings publish financial data as PDF tables. Analysts need this data in spreadsheets for modeling, ratio calculations, and trend analysis. Manual re-entry of a 50-row balance sheet introduces transcription errors and wastes hours of analyst time.
• Academic research — Research papers embed experimental results, survey data, and statistical tables in PDF format. Researchers performing meta-analyses or literature reviews need structured data for aggregation across dozens of source papers.
• Government and regulatory data — Census data, import/export statistics, environmental monitoring reports, and legislative records are published as PDFs. Policy analysts and journalists extract this data to build datasets for public interest reporting and policy evaluation.
• Invoice processing — Accounts payable teams receive invoices as PDFs containing line-item tables with product descriptions, quantities, unit prices, and totals. Extracting these tables feeds automated reconciliation workflows without manual data entry.
• Supply chain documentation — Bills of lading, packing lists, and customs declarations contain tabular shipment data. Logistics teams extract this data to update inventory management and tracking systems.

This tool processes everything locally in your browser using pdfjs-dist. No server receives your document, making it suitable for confidential financial data, privileged legal documents, and any file where privacy is essential.