Skip to content

OCR Stream Parser

OcrStreamParser is designed for scanned or image-heavy PDFs where the text layer is missing, weak, or unusable.

It uses OCR to recover text and then applies stream-style interpretation to build tabular output.

When to use OcrStreamParser

Use OcrStreamParser when:

  • the PDF is a scan
  • text cannot be selected in the PDF viewer
  • the text layer is corrupt or incomplete
  • OCR can recover readable content from the page image

Common examples: - scanned invoices - scanned bank statements - photographed documents - image-heavy archival PDFs

How it works

At a high level:

  1. Render the page to an image
  2. Run OCR on the rendered content
  3. Recover text blocks and positions
  4. Group text into row/column-like structure
  5. Build tables from OCR output
  6. Return List<Table>

This gives you a way to extract tables even when the original PDF has no usable embedded text.

Strengths

  • Works on image-only PDFs
  • Useful for legacy scans
  • Helps recover structure where no text layer exists
  • Good fallback for OCR-readable documents

Limitations

OcrStreamParser depends on OCR quality.

It can degrade when: - the scan is blurry - the page is skewed - text is faint or noisy - the language data is missing - resolution is too low

If visible table borders are strong, LatticeParser may perform better.

Example

import com.extractpdf4j.helpers.Table;
import com.extractpdf4j.parsers.OcrStreamParser;

import java.util.List;

public class OcrStreamExample {
    public static void main(String[] args) throws Exception {
        List<Table> tables = new OcrStreamParser("scan.pdf")
                .pages("1-2")
                .dpi(300f)
                .parse();

        if (!tables.isEmpty()) {
            System.out.println(tables.get(0).toCSV(','));
        }
    }
}

OCR setup notes

For OCR-backed parsing, you may need:

  • Tesseract language data
  • TESSDATA_PREFIX if language data is not automatically found

Example: 

export TESSDATA_PREFIX=/path/to/tessdata

DPI guidance

OCR quality often improves when you render at a better DPI.

Recommended starting point:

  • 300f

For low-quality scans:

  • 400f to 450f

Trade-off:

  • higher DPI may improve recognition
  • but increases processing cost

When OCR Stream is a better choice than Lattice

Choose OcrStreamParser over LatticeParser when:

  • text is readable after OCR
  • borders are weak or absent
  • layout is not strongly ruled
  • you need text-first recovery rather than line-first reconstruction