March 1, 2026 8 min read

    PDF OCR Accuracy Guide: DPI, PSM, OEM and Language Tuning

    Get better OCR output quality for invoices, receipts, and scanned forms.

    1. DPI matters most

    For low-quality scans, increase dpi (200–300 range usually helps). Higher DPI improves recognition but increases processing time.

    2. Tune PSM to document layout

    • Single block invoices: lower segmentation complexity.
    • Multi-column reports: use broader layout parsing.

    3. Set correct language packs

    Use lang matching your document language(s), e.g. eng or eng+hin, to reduce substitution errors.

    curl -X POST https://pdfmunk.com/api/v1/pdf/ocr/parse \
  -H "CLIENT-API-KEY: your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/scanned.pdf",
    "pages": "1-3",
    "lang": "eng",
    "dpi": 240,
    "psm": 3,
    "oem": 3
  }'

    4. Run post-processing

    Normalize whitespace, fix line breaks, and validate known patterns (invoice numbers, totals, dates).

    Conclusion

    DPI + language + segmentation strategy delivers most OCR gains. Iterate on a small golden dataset before scaling. Start from PDF OCR API.

    PDF OCR Accuracy Guide: DPI, PSM, OEM and Language Tuning

    Best practices for OCR quality, preprocessing, language handling, and validation workflows. This page is part of the PDF Munk API platform used for document generation and processing workflows such as HTML to PDF, URL capture, image conversion, OCR, merging, splitting, compression, watermarking, and secure file lifecycle controls.

    Developers typically start with interactive tests, then move the same payloads into backend services, scheduled jobs, and workflow automation tools. You can use this route to validate request structure, evaluate response behavior, and confirm output quality before production rollout.

    Canonical URL: https://pdfmunk.com/blog/pdf-ocr-accuracy-guide. For implementation guidance, review API Docs, run examples in Try Now, and check integration references for n8n and Zapier on the tutorials and blog pages.

    Common production patterns include generating invoices from HTML templates, capturing webpages for legal records, extracting searchable text from scanned files, transforming PDF pages into preview images, and combining or splitting files in approval workflows. Teams often pair these endpoints with queue workers, idempotent retry logic, and structured logging so conversion jobs remain reliable during traffic spikes and downstream API delays.

    When implementing this route, validate input payloads early, keep output mode consistent per workflow, and add monitoring for latency, error rates, and response integrity. For sensitive documents, enforce least-privilege API key handling, rotate credentials periodically, and delete temporary files using lifecycle endpoints once processing is complete. These operational practices improve reliability, security, and cost control as document volume grows.

    Implementation checklist for teams

    Before going live, define request validation rules, decide whether responses should return files or URLs, and set clear retry behavior for network failures. Use consistent timeout values across services, track request IDs end-to-end, and record conversion outcomes for auditing. In batch workflows, split large jobs into smaller units so retries are cheaper and easier to reason about. If you process user-uploaded files, normalize inputs, enforce file-size limits, and surface actionable error messages when payloads are invalid or inaccessible.

    For SEO and rendering quality, keep templates deterministic, pin fonts where possible, and test with representative documents instead of only minimal samples. Add smoke tests for key paths such as create, transform, OCR, and delete operations. If your business depends on predictable output formatting, run visual regression checks on generated documents and store known-good fixtures. These practices reduce operational surprises and help teams maintain stable document automation as APIs, templates, and customer data evolve.

    Need a practical starting point? Begin with a single route, ship observability first, then expand endpoint coverage incrementally. Most teams achieve faster rollout by standardizing request wrappers, centralizing credential handling, and documenting common payload patterns for engineers and no-code operators alike.