Paper scans: creases, shadows, and skew in OCR workflows
Paper scans battle creases, shadows, bleed-through, and skew—harder than crisp UI screenshots. `scan-document-ocr` starts with capture discipline: diffuse lighting, top-down framing, full margins for deskew. Work page by page, watch for dropped line breaks or merged columns in tables, and keep a 300 dpi-class master image when archives matter. OCR text aids search and prep, but for legal-grade retention follow counsel on whether imagery or transcript is authoritative.
Document scan OCR checklist (`scan-document-ocr`)
- Open `scan-document-ocr`, upload each page with gentle perspective correction if you shot it with a phone.
- OCR in sections on long contracts; verify headings, page numbers, and table columns did not collapse together.
- Store the cleaned transcript with page references next to the high-resolution scan package.
Scan OCR FAQ
Why do scanned pages lose the last line or merge table columns?
Shadows and folds break strokes—flatten the sheet, add soft light, and split wide tables into left/right crops before OCR.
Two-column layouts collapse into one paragraph—what is the fastest fix?
OCR the left and right columns separately, then stitch the narrative in reading order; single-pass OCR rarely preserves complex grids.
Page numbers or headers vanish from the transcript—does that hurt search?
Yes. Reinstate headers and page markers manually or encode them in filenames/metadata so future lookups can cite the right clause.
Phone photos show heavy perspective—must you deskew before OCR?
Get as orthogonal as possible first; severe keystone needs perspective correction or smaller frontal crops before text extraction.
Can OCR output replace legal-grade evidence by itself?
Usually not without counsel: keep high-resolution imagery, link it to the transcript, and follow jurisdiction-specific record rules.