33 lines
1.2 KiB
Markdown
33 lines
1.2 KiB
Markdown
|
---
|
||
|
|
name: ocr-document-processor
|
||
|
|
description: Extract text and structure from scans, images, and scanned PDFs. Use for OCR, searchable PDFs, table extraction, receipt parsing, and business card parsing.
|
||
|
|
---
|
||
|
|
|
||
|
|
# OCR Document Processor
|
||
|
|
|
||
|
|
Handle OCR-heavy inputs where text must be recovered from images or scanned pages.
|
||
|
|
|
||
|
|
## Use This For
|
||
|
|
|
||
|
|
- OCR on images and scanned PDFs
|
||
|
|
- Searchable PDF export
|
||
|
|
- Structured extraction to text, markdown, JSON, or HTML
|
||
|
|
- Table extraction from scanned material
|
||
|
|
- Receipt parsing and business card parsing
|
||
|
|
|
||
|
|
## Workflow
|
||
|
|
|
||
|
|
1. Decide whether plain OCR, structured extraction, or document-specific parsing is needed.
|
||
|
|
2. Preprocess noisy inputs before extraction when skew, blur, or shadows are present.
|
||
|
|
3. Use `scripts/ocr_processor.py` for core OCR tasks.
|
||
|
|
4. Use the focused helpers when the input is specialized:
|
||
|
|
- `scripts/business_card_scanner.py`
|
||
|
|
- `scripts/receipt_scanner.py`
|
||
|
|
5. Return confidence caveats when the source is low quality, rotated, handwritten, or multilingual.
|
||
|
|
|
||
|
|
## Guardrails
|
||
|
|
|
||
|
|
- Prefer explicit language selection when accuracy matters.
|
||
|
|
- Do not claim fields are exact when OCR confidence is weak.
|
||
|
|
- Route non-scanned digital PDFs to `document-converter-suite` instead of OCR by default.
|