Optical Character Recognition (OCR)

​

Overview

• Text extraction from images in 70+ languages
• Scanned document processing
• PDF page-by-page OCR
• Multi-format image support
• Indexed and searchable OCR results

​

Powered by Tesseract 5

• Improved accuracy over previous versions
• LSTM neural network-based recognition
• Better handling of complex layouts
• Support for multiple languages simultaneously

// Detected at startup
LOGGER.info("Detected Tesseract " + tessVersion);

​

Supported Image Formats

​

Direct Support (Native Tesseract)

• PNG
• JPEG
• TIFF (multi-page)
• BMP
• PBM, PGM, PPM (Portable formats)

​

Extended Support (with conversion)

• GIF
• WebP
• JPEG 2000 (JP2, JPX)
• HEIC/HEIF
• Adobe Photoshop (PSD)
• RAW camera formats
• SVG, EMF, WMF vector formats

​

Non-Image Formats

• PDF - Each page converted to image for OCR

​

Configuration

# Enable OCR processing
tesseract.enabled = true

# Tesseract binary path (leave empty if in PATH)
tesseract.path = 

# Language model (por, eng, spa, etc.)
ocr.language = por

# Page segmentation mode (1-13)
ocr.pageSegMode = 1

# Skip files found in hash database
ocr.skipKnownFiles = false

# File size limits (bytes)
ocr.minFileSize = 10000
ocr.maxFileSize = 100000000

# Process non-standard image formats
ocr.processNonStandard = true

# Maximum image dimension for conversion
ocr.maxConvImageSize = 3000

​

Language Models

​

Pre-installed

• Portuguese (por)
• English (eng)

​

Additional Languages

• Spanish (spa)
• French (fra)
• German (deu)
• Italian (ita)
• Chinese (chi_sim, chi_tra)
• Arabic (ara)
• Russian (rus)
• Japanese (jpn)
• And 60+ more languages

​

Page Segmentation Modes

• 0 - Orientation and script detection only
• 1 - Automatic page segmentation with OSD (default)
• 3 - Fully automatic without OSD
• 4 - Single column of text
• 6 - Single uniform block of text
• 7 - Single text line
• 11 - Sparse text (find as much text as possible)
• 13 - Raw line (treats image as single text line)

​

Implementation Details

​

1. File Filtering

if (size >= MIN_SIZE && size <= MAX_SIZE 
    && !SKIP_KNOWN_FILES && itemInfo.isKnown()) {
    // Process OCR
}

​

2. Image Preparation

• Read directly into Tesseract

private void parsePDF(XHTMLContentHandler xhtml, File input, File output) {
    PDFToImage pdfConverter = new PDFToImage();
    pdfConverter.load(input);
    for (int page = 0; page < pdfConverter.getNumPages(); page++) {
        // Convert page to image
        // OCR the image
        // Aggregate results
    }
}

private void parseTiff(XHTMLContentHandler xhtml, File input) {
    ImageReader reader = ImageIO.getImageReaders(iis).next();
    int numPages = reader.getNumImages(true);
    // OCR each page separately to avoid timeouts
}

• Convert to standard format using ImageMagick or internal converters
• Resize if exceeds maximum dimensions
• OCR converted image

​

3. Tesseract Execution

String[] cmd = { 
    tesseractPath, 
    INPUT_FILE_TOKEN,   // Input image
    OUTPUT_FILE_TOKEN,  // Output text file
    "-l", LANGUAGE,      // Language model
    "--psm", PAGESEGMODE // Page segmentation mode
};

Process process = new ProcessBuilder(cmd).start();

​

4. Result Storage

CREATE TABLE IF NOT EXISTS ocr(
    id TEXT PRIMARY KEY,     -- Hash + page number
    text TEXT                -- Extracted text
);

• Fast lookup for duplicate files
• Persistent across processing sessions
• Reduced reprocessing time
• Efficient storage

​

Performance Optimization

​

Image Preprocessing

• Downsampling - Large images resized to 3000x3000 max
• Format conversion - Non-standard formats converted once
• Page-by-page - Large documents processed incrementally

​

Multithreading

• OpenMP disabled to prevent thread conflicts:

env.put("OMP_THREAD_LIMIT", "1");

• Each file processed in separate thread by IPED task manager

​

Caching

• Results cached in database by file hash
• Duplicate files skip OCR entirely
• Significant speedup for datasets with duplicates

​

Timeouts

• Configurable timeout per operation
• Automatic retry with conversion for corrupted images

​

Subset Processing

subsetToOcr = Important_#_Bookmarked

• Faster initial processing
• Two-pass workflow (process all, then OCR subset)
• Resource management for large cases

​

Quality Metrics

metadata.set(OCRParser.OCR_CHAR_COUNT, Integer.toString(charCount));

• Character count stored as metadata
• Helps identify successful vs. failed OCR
• Useful for quality assessment

​

Integration Features

​

Indexed Text

• OCR text added to full-text index
• Searchable like native document text
• Hit highlighting in OCR content

​

Result Display

• OCR text shown in text viewer
• Formatted as plain text
• Clearly marked as OCR content

​

Export

• OCR text included in reports
• Available in HTML and CSV exports
• Preserved in portable cases

​

Use Cases

​

Scanned Documents

• Scanned contracts and agreements
• Historical documents
• Medical records
• Financial statements

​

Screenshots

• Chat application screenshots
• Web page captures
• Email screenshots
• Error messages and logs

​

Photo Evidence

• License plates
• Street signs
• Document photographs
• Whiteboard images

​

Memes and Graphics

• Image macros with overlaid text
• Infographics
• Social media content

​

Accuracy Considerations

​

Factors Affecting Accuracy

• Resolution (300+ DPI recommended)
• Contrast and brightness
• Focus and sharpness

• Font type and size
• Text orientation
• Language and script

• Noise and artifacts
• Complex backgrounds
• Skew and rotation

​

Improving Results

1. Choose appropriate language model - Match document language
2. Adjust page segmentation - Try different modes for layout types
3. Preprocess images - Enhance contrast, deskew, denoise
4. Use higher DPI - Scan documents at 300 DPI or higher

​

Troubleshooting

​

Low Accuracy

• Verify correct language model selected
• Check image quality and resolution
• Try different page segmentation modes
• Preprocess images to enhance text

​

Performance Issues

• Reduce maxFileSize to skip very large images
• Enable skipKnownFiles to avoid processing known content
• Use subsetToOcr for selective processing
• Increase processing threads if CPU underutilized

​

Missing Text

• Verify OCR is enabled in configuration
• Check file size within min/max limits
• Ensure Tesseract installed and in PATH
• Review OCR character count in metadata

​

External Dependencies

• Tesseract 5.x - OCR engine binary

• ImageMagick - Extended format support
• Ghostscript - PDF to image conversion
• libgif, libopenjp2, libwebp - Additional image format libraries