Supported Formats

Complete reference for document formats supported by GateFlow OCR.

Overview

GateFlow OCR supports a wide range of document formats for text extraction, with specialized handling for complex layouts, tables, and multi-page documents.

Document Formats

PDF Documents

Format	Extension	OCR Required	Notes
Native PDF	.pdf	No	Text directly extractable
Scanned PDF	.pdf	Yes	Image-based, requires OCR
Mixed PDF	.pdf	Partial	Both native and scanned pages
PDF/A	.pdf	Varies	Archival format, full support
Encrypted PDF	.pdf	No*	Requires password

from openai import OpenAI

client = OpenAI(
    base_url="https://api.gateflow.ai/v1",
    api_key="gw_prod_..."
)

# Process a PDF document
response = client.post(
    "/data/ocr",
    files={"file": open("document.pdf", "rb")},
    data={
        "detect_type": True,  # Auto-detect native vs scanned
        "extract_tables": True,
        "preserve_layout": True
    }
)

print(f"Document type: {response['document_type']}")
print(f"Pages: {response['page_count']}")
print(f"OCR applied: {response['ocr_applied']}")

Image Formats

Format	Extension	Best For	Max Resolution
JPEG	.jpg, .jpeg	Photos, scans	10000x10000
PNG	.png	Screenshots, diagrams	10000x10000
TIFF	.tif, .tiff	High-quality scans	10000x10000
WebP	.webp	Web images	10000x10000
BMP	.bmp	Legacy scans	10000x10000
HEIC	.heic	iOS photos	10000x10000

# Process an image
response = client.post(
    "/data/ocr",
    files={"file": open("scan.jpg", "rb")},
    data={
        "deskew": True,       # Correct rotation
        "denoise": True,      # Remove noise
        "enhance_contrast": True
    }
)

Office Documents

Format	Extension	Native Text	OCR for Images
Word	.docx	Yes	Embedded images
Word (Legacy)	.doc	Yes	Embedded images
Excel	.xlsx	Yes	Embedded images
Excel (Legacy)	.xls	Yes	Embedded images
PowerPoint	.pptx	Yes	Slides as needed
PowerPoint (Legacy)	.ppt	Yes	Slides as needed

# Process a Word document with embedded images
response = client.post(
    "/data/ocr",
    files={"file": open("report.docx", "rb")},
    data={
        "ocr_embedded_images": True,
        "preserve_formatting": True
    }
)

Plain Text and Markup

Format	Extension	Processing
Plain Text	.txt	Direct extraction
Markdown	.md	Structure preserved
HTML	.html, .htm	Tags stripped, structure kept
XML	.xml	Content extraction
CSV	.csv	Table structure
JSON	.json	Data extraction

Email Formats

Format	Extension	Attachments
EML	.eml	Processed recursively
MSG	.msg	Processed recursively
MBOX	.mbox	Multiple emails

# Process email with attachments
response = client.post(
    "/data/ocr",
    files={"file": open("email.eml", "rb")},
    data={
        "process_attachments": True,
        "max_attachment_depth": 2,
        "attachment_types": ["pdf", "docx", "jpg"]
    }
)

print(f"Email body extracted")
print(f"Attachments processed: {response['attachments_count']}")

Image Quality Requirements

Minimum Requirements

Parameter	Minimum	Recommended	Optimal
Resolution	150 DPI	300 DPI	600 DPI
Text height	8px	20px	32px+
Contrast ratio	2:1	4:1	7:1+
Skew angle	< 15°	< 5°	< 1°

Quality Preprocessing

# Enable automatic quality enhancement
response = client.post(
    "/data/ocr",
    files={"file": open("poor_scan.jpg", "rb")},
    data={
        "preprocessing": {
            "deskew": True,
            "denoise": True,
            "binarize": True,
            "remove_borders": True,
            "enhance_resolution": True  # AI upscaling
        }
    }
)

print(f"Quality score before: {response['quality']['original']}")
print(f"Quality score after: {response['quality']['enhanced']}")

Language Support

Supported Languages

GateFlow OCR supports 100+ languages including:

Category	Languages
Latin	English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Czech, etc.
Cyrillic	Russian, Ukrainian, Bulgarian, Serbian, etc.
Asian	Chinese (Simplified & Traditional), Japanese, Korean, Thai, Vietnamese
Middle Eastern	Arabic, Hebrew, Persian, Urdu
South Asian	Hindi, Bengali, Tamil, Telugu, Gujarati
Other	Greek, Armenian, Georgian, etc.

# Specify language for better accuracy
response = client.post(
    "/data/ocr",
    files={"file": open("german_doc.pdf", "rb")},
    data={
        "languages": ["de", "en"],  # Primary and secondary
        "detect_language": True
    }
)

print(f"Detected languages: {response['detected_languages']}")

Multi-Language Documents

# Process document with multiple languages
response = client.post(
    "/data/ocr",
    files={"file": open("multilingual.pdf", "rb")},
    data={
        "languages": ["en", "es", "fr", "de"],
        "per_page_detection": True
    }
)

for page in response["pages"]:
    print(f"Page {page['number']}: {page['detected_language']}")

File Size Limits

Plan	Max File Size	Max Pages	Max Resolution
Free	10 MB	10 pages	2000x2000
Pro	50 MB	100 pages	5000x5000
Enterprise	500 MB	1000 pages	10000x10000

Handling Large Documents

# Process large document in chunks
response = client.post(
    "/data/ocr",
    files={"file": open("large_doc.pdf", "rb")},
    data={
        "page_range": "1-100",  # First 100 pages
        "async": True           # Process asynchronously
    }
)

job_id = response["job_id"]

# Check status
status = client.get(f"/data/ocr/jobs/{job_id}")
print(f"Status: {status['status']}")
print(f"Progress: {status['pages_processed']}/{status['total_pages']}")

Output Formats

Text Output

# Get plain text output
response = client.post(
    "/data/ocr",
    files={"file": open("document.pdf", "rb")},
    data={"output_format": "text"}
)

print(response["text"])

Structured Output

# Get structured output with layout
response = client.post(
    "/data/ocr",
    files={"file": open("document.pdf", "rb")},
    data={"output_format": "structured"}
)

for page in response["pages"]:
    print(f"\n--- Page {page['number']} ---")
    for block in page["blocks"]:
        print(f"[{block['type']}] {block['text'][:100]}...")

JSON Output

# Get full JSON output
response = client.post(
    "/data/ocr",
    files={"file": open("document.pdf", "rb")},
    data={
        "output_format": "json",
        "include_coordinates": True,
        "include_confidence": True
    }
)

# Response includes bounding boxes and confidence scores
for word in response["pages"][0]["words"]:
    print(f"{word['text']} ({word['confidence']:.2f})")
    print(f"  Position: {word['bbox']}")

Markdown Output

# Get Markdown output preserving structure
response = client.post(
    "/data/ocr",
    files={"file": open("document.pdf", "rb")},
    data={
        "output_format": "markdown",
        "preserve_headings": True,
        "preserve_lists": True,
        "preserve_tables": True
    }
)

print(response["markdown"])

Error Handling

Common Format Errors

Error	Cause	Solution
unsupported_format	File type not supported	Convert to supported format
corrupted_file	File is damaged	Re-upload or repair file
password_protected	PDF requires password	Provide password parameter
too_large	Exceeds size limit	Split document or upgrade plan
low_quality	Image quality too poor	Enhance or rescan

# Handle format errors
try:
    response = client.post(
        "/data/ocr",
        files={"file": open("document.pdf", "rb")}
    )
except Exception as e:
    if "password_protected" in str(e):
        # Retry with password
        response = client.post(
            "/data/ocr",
            files={"file": open("document.pdf", "rb")},
            data={"password": "secret123"}
        )

Best Practices

1. Use native PDFs when possible - Faster and more accurate than OCR
2. Scan at 300 DPI minimum - Better accuracy for OCR
3. Specify languages - Improves recognition accuracy
4. Enable preprocessing - Helps with poor quality scans
5. Use async for large files - Prevents timeouts
6. Check confidence scores - Flag low-confidence text for review

Next Steps

• Mistral Document AI - OCR engine details
• Table Extraction - Extract tables
• Document Ingestion - Full pipeline