Document OCR & PDF Text API

Extract text from public PDF and image URLs with a hybrid pipeline optimized for both quality and cost.

For PDFs, the Actor first checks each page for a usable embedded text layer. If the page already contains real text, the Actor extracts it directly with Poppler pdftotext. If usable text is not available, the page is rendered and processed with local Tesseract OCR. Raster image inputs always use Tesseract OCR.

This avoids wasting OCR compute on text-native PDFs while preserving OCR fallback for scanned PDFs and images.

Input

Provide one or more public http:// or https:// URLs in documentUrls.

Use:

• maxPagesPerDocument to bound PDF pages,
• maxResults to bound total emitted page rows,
• maxDownloadMb to cap each source download,
• language for the installed OCR language,
• skipInvalidUrls to continue after blocked or invalid sources.

Security

The Actor is intended only for public internet documents. It blocks URL credentials, localhost, loopback, private, link-local, reserved/non-global IP addresses, and common metadata hostnames. Redirect targets are validated again before they are followed. Downloads are byte-limited.

Output

Each Dataset row represents one page or image and includes extractionMethod:

• pdf-text — usable text was extracted directly from the PDF page.
• tesseract-ocr — raster OCR was required.

The Actor never emits packaged benchmark fixtures in customer runs. Only user-provided source URLs create Dataset rows.

OCR behavior

Tesseract is strongest on clear machine-printed text. Handwriting, poor scans, perspective distortion, unusual scripts, complex forms, and heavy compression can reduce accuracy.

The Actor does not claim universal OCR accuracy. High-impact legal, medical, financial, compliance, or identity use should independently verify extracted text.

Billing

The primary pay-per-event event is page-result. One result event corresponds to one page or image successfully written to the Dataset. Billing verification uses the event-specific charging counter.

Example

{
  "documentUrls": [
    "https://www.irs.gov/pub/irs-pdf/fw9.pdf"
  ],
  "maxPagesPerDocument": 6,
  "maxResults": 20,
  "maxDownloadMb": 30,
  "language": "eng",
  "skipInvalidUrls": true
}