Sensitive information in images and scans: Why OCR belongs in your DLP evaluation

Quick answer

Include image-only documents, screenshots and scans of different quality. Evaluate extracted text separately from policy outcomes rather than treating text-file results as evidence for image handling.

Text search may miss details embedded in images, document scans, and screenshots. Organizations relying on scanned documents should include them in the original DLP scope rather than testing only spreadsheets.

Verify capabilities and limitations

Endpoint DLP Plus describes OCR-based image scanning. Thai-language accuracy, document layout, and image quality must be tested in the actual version. Product-page demonstration figures are not evidence of your organization's detection rate. Check supported formats and size constraints.

Build representative synthetic examples

Include image-only PDFs, angled scans, and mixed Thai-English screenshots with invented information. Separate sensitive examples from ordinary content. Record text-recognition outcomes and policy decisions independently to identify the cause of a miss.

Handle unreadable documents explicitly

Ask data owners to assign handling rules and alternative classification measures for documents that cannot be reliably processed. Improve scanning practices where appropriate. Failure to extract text does not establish that a file is non-confidential.

Implementation checklist

• Vary language, layout, and image quality in synthetic fixtures.
• Include ordinary documents to assess false positives.
• Record unsupported formats and processing failures.
• Retest after version or detection-rule changes.

A useful pilot records detections, misses, unsupported cases, and compensating measures. KTNBS can prepare an evaluation using synthetic documents representative of your workflows.

A scenario to take into your evaluation

Use identical synthetic text in a sharp image, skewed scan and image-only PDF; inspect extraction and alerts.

This is a proposed test, not a measured result or customer endorsement. Record the actual platform and edition, owner, expected result and observed outcome before expanding scope.