PDF Text Extraction: Pdf.ExtractText

The operations

The op argument selects what comes back in jsonOut. Every operation also returns the quality object. Returns true on success; a page with no extractable text (a scanned image) is still success, and the quality report says why. Chilkat does not perform OCR.

Quick start

An empty jsonIn extracts every page in automatic reading order, without artifacts, with the standard clean-up.

Chilkat.Pdf pdf = new Chilkat.Pdf();
pdf.LoadFile("report.pdf");

Chilkat.JsonObject jsonIn = new Chilkat.JsonObject();
Chilkat.JsonObject jsonOut = new Chilkat.JsonObject();

if (!pdf.ExtractText("text", jsonIn, jsonOut)) {
    Console.WriteLine(pdf.LastErrorText);
    return;
}
string text = jsonOut.StringOf("text");            // all pages, joined
string rating = jsonOut.StringOf("quality.rating"); // good, fair, poor or none

{ }

{
  "text": "Quarterly Report\nRevenue grew 12% ...\f...",
  "pages": [
    { "page": 1, "text": "Quarterly Report\nRevenue grew 12% ..." },
    { "page": 2, "text": "..." }
  ],
  "quality": {
    "rating": "good",
    "tagged": true,
    "structureUsable": true,
    "pagesNeedingOcr": "",
    "pages": [ ... ]
  }
}

Scope: pages and labels (all operations)

Scope is set by page and, optionally, by a label in the text.

The label scope works with every operation: text returns the value, markdown returns it as Markdown, search looks only inside the labelled text, and tables returns only the tables inside it (the line-item table under "Items", for example).

Example: capture a field by its label

{ "label": "Invoice Number" }

{
  "text": "INV-2026-0412",
  "pages": [ { "page": 1, "text": "INV-2026-0412" } ],
  "labelMatches": [
    { "page": 1, "label": "Invoice Number:", "text": "INV-2026-0412" }
  ],
  "quality": { "rating": "good", ... }
}

Example: an address block under a label

{ "label": "Bill To", "labelValue": "below" }

{
  "text": "Acme Corp\n100 Main Street\nAustin, TX 78701",
  "pages": [ { "page": 1, "text": "Acme Corp\n100 Main Street\nAustin, TX 78701" } ],
  "labelMatches": [
    { "page": 1, "label": "Bill To", "text": "Acme Corp\n100 Main Street\nAustin, TX 78701" }
  ],
  "quality": { ... }
}

Example: every occurrence, on selected pages, by regular expression

{
  "pages": "2-5",
  "label": "(Phone|Tel)\\.?",
  "labelRegex": true,
  "labelValue": "right",
  "labelOccurrence": "all"
}

{
  "text": "866 607 7339\n852 3160 9833\n44 (0) 1440 711 822",
  "pages": [ { "page": 2, "text": "866 607 7339\n852 3160 9833\n44 (0) 1440 711 822" } ],
  "labelMatches": [
    { "page": 2, "label": "Phone:", "text": "866 607 7339" },
    { "page": 2, "label": "Phone:", "text": "852 3160 9833" },
    { "page": 2, "label": "Phone:", "text": "44 (0) 1440 711 822" }
  ],
  "quality": { ... }
}

Reading order (all operations)

{ "readingOrder": "visual", "columns": "none", "layout": "lines" }

{
  "text": "...",
  "pages": [ { "page": 1, "text": "..." }, ... ],
  "quality": { "rating": "good", ...,
    "pages": [ { "page": 1, "readingOrderSource": "visual", "columns": 1, ... }, ... ] }
}

{ "readingOrder": "structure" }

{
  "text": "...",
  "pages": [ ... ],
  "quality": { "rating": "good", "tagged": true, "structureUsable": true, ...,
    "pages": [ { "page": 1, "readingOrderSource": "structure", ... }, ... ] }
}

Filters (all operations)

Each filter is include or exclude and decides whether a kind of text appears in the output.

{
  "artifacts": "include",
  "repeatedHeadersFooters": "include",
  "invisibleText": "include",
  "clippedText": "include",
  "offPageText": "include",
  "obscuredText": "include",
  "formFields": "include",
  "annotations": "include"
}

{
  "text": "...",
  "pages": [ ... ],
  "quality": { "rating": "fair", ...,
    "pages": [ { "page": 2, "textSource": "mixed",
                 "issues": [ { "code": "rotatedText", "message": "Rotated text was included" } ], ... } ] }
}

{ "formFields": "include", "layout": "lines" }

{
  "text": "Name: Jane Doe\nAccount: 44-1029\nSignature: ...",
  "pages": [ { "page": 1, "text": "Name: Jane Doe\nAccount: 44-1029\n..." } ],
  "quality": { ... }
}

Clean-up (all operations)

{
  "expandLigatures": false,
  "dehyphenate": false,
  "actualText": false,
  "normalize": "none",
  "lineEnding": "\r\n"
}

{
  "text": "The ligature \ufb01le and the hyphen-\r\nated word stay as drawn...",
  "pages": [ ... ],
  "quality": { ... }
}

With lineEnding given as a JSON string, write it as "\r\n" in the JSON text (the JSON escape for a carriage return and a line feed), or simply "crlf".

op = text

Results

Example: the three layouts of the same invoice line

{ "pages": "1", "layout": "grid", "pageSeparator": "\n\n" }

flow:   Widget, blue 2 $10.00 $20.00 Subtotal $20.00
lines:  Widget, blue    2    $10.00    $20.00
        Subtotal                       $20.00
grid:   Widget, blue          2      $10.00     $20.00
        Subtotal                                $20.00

op = markdown

Results

markdown (all pages joined, with the page breaks chosen), pages[i].page / pages[i].markdown, labelMatches[i].markdown when label is set, and quality.

Example: Markdown for an AI pipeline, with a quality gate

pdf.ExtractText("markdown", jsonIn, jsonOut);
string md = jsonOut.StringOf("markdown");
if (jsonOut.StringOf("quality.rating") == "poor") {
    // send for review, or OCR the pages listed in quality.pagesNeedingOcr
}

{
  "pageBreaks": "rule",
  "images": "none",
  "footnotes": "end",
  "maxHeadingLevel": 2
}

# Quarterly Report

## Revenue

Revenue grew **12%** over the prior quarter, driven by
the *Widget* line.[^1]

| Product | Units | Revenue |
| --- | --- | --- |
| Widget, blue | 2 | $20.00 |
| Widget, red | 5 | $50.00 |

See [the appendix](https://example.com/appendix) for details.

---

## Outlook

- Expand the blue line
- Open the Austin office

[^1]: Excluding returns.

Example: one section as Markdown

{ "label": "Terms and Conditions", "labelValue": "section" }

{
  "markdown": "## Terms and Conditions\n\nPayment is due within 30 days ...",
  "pages": [ { "page": 4, "markdown": "## Terms and Conditions\n\n..." } ],
  "labelMatches": [ { "page": 4, "label": "Terms and Conditions", "markdown": "## Terms and Conditions\n\n..." } ],
  "quality": { ... }
}

op = search

Results

count, and matches[i] with page, text (the matched text), before and after (the context), plus quality.

Example: find an invoice total with a regular expression

jsonIn.UpdateString("query", @"Total\s+\$?[0-9,]+\.[0-9]{2}");
jsonIn.UpdateBool("regex", true);
pdf.ExtractText("search", jsonIn, jsonOut);
string total = jsonOut.StringOf("matches[0].text");
int page = jsonOut.IntOf("matches[0].page");

{
  "query": "Total\\s+\\$?[0-9,]+\\.[0-9]{2}",
  "regex": true,
  "context": 20
}

{
  "count": 2,
  "matches": [
    { "page": 1, "text": "Total $1,234.56",
      "before": "   $1,234.56\nSub", "after": "\nThank you for your" },
    { "page": 3, "text": "Total $98.00",
      "before": "Shipping   $8.00\n", "after": "\nPaid by card" }
  ],
  "quality": { ... }
}

Example: a plain word, whole words only, first 10 hits

{ "query": "warranty", "wholeWord": true, "maxResults": 10 }

{
  "count": 7,
  "matches": [
    { "page": 2, "text": "warranty", "before": "This limited ", "after": " covers defects in" },
    ...
  ],
  "quality": { ... }
}

{ "label": "Exclusions", "labelValue": "section", "query": "water damage" }

{
  "count": 1,
  "matches": [
    { "page": 5, "text": "water damage",
      "before": "not covered: ", "after": ", misuse, or" }
  ],
  "quality": { ... }
}

op = tables

Results

count, and tables[i] with page, detectedBy (ruled, aligned or structure), rows, columns, and cells[j] with row, col (counted from 0), rowSpan, colSpan and text.

{ "pages": "1", "minRows": 2, "minColumns": 3 }

{
  "count": 1,
  "tables": [
    { "page": 1, "detectedBy": "ruled", "rows": 3, "columns": 3,
      "cells": [
        { "row": 0, "col": 0, "rowSpan": 1, "colSpan": 1, "text": "Product" },
        { "row": 0, "col": 1, "rowSpan": 1, "colSpan": 1, "text": "Units" },
        { "row": 0, "col": 2, "rowSpan": 1, "colSpan": 1, "text": "Revenue" },
        { "row": 1, "col": 0, "rowSpan": 1, "colSpan": 1, "text": "Widget, blue" },
        { "row": 1, "col": 1, "rowSpan": 1, "colSpan": 1, "text": "2" },
        { "row": 1, "col": 2, "rowSpan": 1, "colSpan": 1, "text": "$20.00" },
        { "row": 2, "col": 0, "rowSpan": 1, "colSpan": 2, "text": "Total" },
        { "row": 2, "col": 2, "rowSpan": 1, "colSpan": 1, "text": "$20.00" }
      ] }
  ],
  "quality": { ... }
}

int n = jsonOut.SizeOfArray("tables[0].cells");
for (int i = 0; i < n; i++) {
    jsonOut.I = i;
    Console.WriteLine(jsonOut.IntOf("tables[0].cells[i].row") + "," + jsonOut.IntOf("tables[0].cells[i].col")
        + ": " + jsonOut.StringOf("tables[0].cells[i].text"));
}

op = assess

assess returns only the quality report. It takes the same scope, reading-order and filter options as the other operations (they decide what is assessed), never returns text, and is the one operation allowed on an encrypted PDF that forbids copying.

pdf.ExtractText("assess", jsonIn, jsonOut);
string ocrPages = jsonOut.StringOf("quality.pagesNeedingOcr");   // e.g. "4,9"
bool tagged = jsonOut.BoolOf("quality.tagged");

The quality report

Every operation returns a quality object that rates each page and the whole document and names the reasons, so an application can decide whether to trust the text, send pages to OCR, or flag the output for review.

"quality": {
  "rating": "fair",
  "tagged": true,
  "structureUsable": false,
  "pagesNeedingOcr": "4,9",
  "pages": [
    { "page": 3, "rating": "fair",
      "textSource": "text",
      "unicodeCoverage": 0.97,
      "unmappedFonts": [ "F12+Wingdings" ],
      "suspiciousCharRatio": 0.004,
      "readingOrderSource": "visual",
      "columns": 2,
      "structure": { "headings": "inferred", "tables": { "ruled": 1, "aligned": 0 } },
      "issues": [
        { "code": "multiColumn", "message": "Multi-column layout; reading order inferred" },
        { "code": "missingUnicode", "message": "Some glyphs have no reliable Unicode mapping" } ] },
    { "page": 4, "rating": "none", "textSource": "imageOnly",
      "unicodeCoverage": 0, "suspiciousCharRatio": 0,
      "readingOrderSource": "visual", "columns": 0,
      "structure": { "headings": "none", "tables": { "ruled": 0, "aligned": 0 } },
      "issues": [ { "code": "imageOnly", "message": "No text; page is a scanned image" } ] }
  ]
}

ExtractTextSb: large documents

ExtractTextSb(op, jsonIn, sb, jsonOut) is the same call, except that the joined text or Markdown is written into the StringBuilder instead of the text / markdown member of jsonOut. Everything else (pages, labelMatches, search and table results, quality) is still returned in jsonOut.

Chilkat.StringBuilder sb = new Chilkat.StringBuilder();
if (pdf.ExtractTextSb("markdown", jsonIn, sb, jsonOut)) {
    sb.WriteFile("report.md", "utf-8", false);
}

Permissions and signed PDFs

• Extraction is read-only: it never affects signatures, pending changes, or what WriteFile writes.
• For an encrypted PDF, every operation except assess requires owner rights or the copy permission. Chilkat does not ignore owner restrictions; load the PDF with the OwnerPassword property set to extract from a PDF that restricts copying. The failure reason is in LastErrorText.
• An encrypted PDF that could not be opened (wrong password) refuses every operation except assess.
• Watermarks added by Chilkat or Acrobat are excluded by the artifacts filter, the same recognition that Watermark detect uses.

Option reference

Every member of jsonIn, grouped as above. All of them are optional except query for search.

{
  "pages": "all",                       "label": "",            "labelValue": "auto",
  "labelOccurrence": "first",           "labelRegex": false,
  "readingOrder": "auto",               "columns": "auto",
  "artifacts": "exclude",               "repeatedHeadersFooters": "exclude",
  "invisibleText": "auto",              "clippedText": "exclude",
  "offPageText": "exclude",             "obscuredText": "exclude",
  "rotatedText": "include",             "formFields": "exclude",
  "annotations": "exclude",
  "expandLigatures": true,              "dehyphenate": true,
  "actualText": true,                   "bidi": true,
  "normalize": "NFC",                   "lineEnding": "\n",
  "layout": "flow",                     "pageSeparator": "\f",            // text
  "headings": true,                     "maxHeadingLevel": 3,             // markdown
  "emphasis": true,                     "lists": true,
  "links": true,                        "code": true,
  "tables": "ruled",                    "images": "placeholder",
  "pageBreaks": "comment",              "footnotes": "inline",
  "query": "",                          "regex": false,                   // search
  "caseSensitive": false,               "wholeWord": false,
  "maxResults": 0,                      "context": 40,
  "detection": "ruled",                 "minRows": 2,                     // tables
  "minColumns": 2
}

Reference documentation: Pdf.ExtractText, Pdf.ExtractTextSb. Related: TextReplace, Watermark, Protection, Optimize, Pdf class overview.