doc-scan（文档扫描）

Doc-Scan — DEPRECATED

This skill is deprecated. All functionality has been merged into doc-process v4.0.0 with a significantly improved scanner engine. Please use doc-process instead. To scan a document photo: install doc-process and say "scan this photo", "correct perspective", "dewarp this document", or any equivalent phrase.

Doc-Scan — Document Scanner Skill (archived)

Converts a photo of a document (whiteboard, printed page, handwritten note, form, receipt, book page, etc.) into a clean scanned-looking image with perspective correction and enhancement.

Step 1 — Validate the Input

Read the provided image visually. Assess:

Non-Document Detection

If the image does not appear to contain a document, respond:

⚠ This image doesn't appear to contain a document.

I detected: [brief description of what the image contains — e.g., "a landscape photo", "a person's portrait", "a blank wall"]

Doc-Scan works best with:
- Printed documents (forms, letters, reports)
- Handwritten notes or whiteboards
- Receipts, invoices, or business cards
- Book or magazine pages
- Any flat document photographed from above or at an angle

If you intended to upload a document photo, please try again with better lighting and the document clearly visible. If you want to process this image for a different purpose, I can help with that instead.

Do not proceed with scanning if this check fails.

Step 2 — Pre-Scan Assessment

Report what you see before scanning:

## Document Photo Assessment

| Property | Detected Value |
|---|---|
| Document type | [e.g., Printed letter, handwritten note, receipt, form] |
| Orientation | Portrait / Landscape / Tilted (~N degrees) |
| Perspective distortion | None / Mild / Moderate / Severe |
| Lighting | Even / Uneven (shadow on [region]) / Too dark / Too bright |
| Background | White desk / Dark table / Complex background |
| Image quality | Sharp / Slightly blurred / Blurred |
| Estimated document area | ~N% of total image |
| Multi-page? | Single page / [N] pages detected |
| Content visible | [brief description — e.g., "text document, 3 columns, appears to be a form"] |

Recommended enhancements:
- [x] Perspective correction
- [x] Background removal / edge crop
- [ ] Binarization (black & white) — suitable if text-only
- [x] Contrast enhancement
- [x] Shadow removal
- [ ] Color preservation — suitable for documents with color content

Step 3 — Run the Scanner Script

python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png

Common Options

# Black and white output (best for text documents)
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png --mode bw

# Color-preserved output (best for forms, diagrams, colored content)
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png --mode color

# Grayscale output (middle ground)
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png --mode gray

# Output as PDF
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.pdf --format pdf

# Multiple images into one PDF (multi-page scan)
python skills/doc-scan/scripts/doc_scanner.py --input page1.jpg page2.jpg page3.jpg --output document.pdf --format pdf

# Manual corner specification (if auto-detection fails)
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png --corners "50,30 800,20 820,1100 40,1120"

# High-resolution output
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png --dpi 300

# Skip perspective correction (if photo is already flat)
python skills/doc-scan/scripts/doc_scanner.py --input photo.jpg --output scanned.png --no-warp

Step 4 — Interpret Script Output

The script outputs a JSON status block to stderr. Parse and report to user:

{
  "status": "success",
  "corners_detected": true,
  "corners": [[50,30],[800,20],[820,1100],[40,1120]],
  "warp_applied": true,
  "enhancement_mode": "bw",
  "input_size": [3024, 4032],
  "output_size": [2480, 3508],
  "output_dpi": 300,
  "pages": 1,
  "output_file": "scanned.png",
  "warnings": []
}

Status Handling

"status": "success": Report completion with key stats.

"corners_detected": false: Auto-detection failed. Offer:

• "Auto edge-detection could not find the document corners. I can try with manual corner hints — please describe approximately where the four corners of the document appear in the photo (e.g., top-left at about 10% from left and 5% from top)."
• Or: offer --no-warp mode to at least apply enhancement without perspective correction

warnings array: Report any warnings to user — e.g., "Low contrast image", "Detected significant blur", "Partial document visible"

Step 5 — Post-Scan Quality Check

After the script completes, read the output image visually and assess:

If any check fails, report the issue and offer:

• Re-run with different settings (different mode, manual corners, contrast level)
• Re-photograph tips (see Step 7)

Step 6 — Output Report

## Scan Complete ✓

| Property | Value |
|---|---|
| Output file | scanned.png |
| Output size | 2480 × 3508 px (A4 at 300 DPI) |
| Mode | Black & White |
| Perspective correction | Applied |
| Shadow removal | Applied |
| Processing time | ~2.3s |

### Enhancements Applied
- Edge detection and four-corner extraction
- Perspective warp to standard A4 dimensions
- Adaptive thresholding (Sauvola method) for clean B&W text
- Shadow compensation via background normalization
- Border cropped to document edges

### Before → After
[Original photo] → [Scanned output]
(Both are available at their file paths)

Step 7 — Multi-Page Documents

If the user provides multiple photos (or a folder of images):

1. Process each image individually
2. Sort by filename or user-specified order
3. Combine into a single PDF:

python skills/doc-scan/scripts/doc_scanner.py \
  --input page1.jpg page2.jpg page3.jpg \
  --output document.pdf --format pdf

1. Report: "3-page document scanned and saved to document.pdf"

For very large batches (10+ pages), process in batches of 10 and combine.

Step 8 — Photography Tips (on poor quality input)

If the input image quality is poor or auto-detection fails, provide tips:

### Tips for Better Scans

**Lighting:**
- Scan in bright, even lighting (avoid direct sunlight creating glare)
- Avoid shadows from your hand or body
- A well-lit indoor environment works well

**Camera position:**
- Hold the camera directly above the document when possible
- Keep the camera parallel to the document surface
- The full document should be visible with a small border around it

**Background:**
- Place the document on a contrasting background (dark table for white paper, white surface for dark paper)
- Avoid patterned or busy backgrounds

**Focus & stability:**
- Tap to focus on the document before taking the photo
- Hold steady or use a tripod for sharper results
- Use the highest resolution available on your camera

**Document preparation:**
- Flatten any folds or curves
- If the document is creased, press it flat before photographing

Step 9 — Integration with Doc-Process

After a successful scan, offer to process the scanned document with doc-process:

Scan complete! Would you also like me to:
- Extract text and fill a form from this scanned document?
- Analyze this as a contract/receipt/medical document?
- Translate this document to another language?
- Extract tables from this document?

Just say what you'd like to do and I'll process the scanned version.

If the user confirms, pass the scanned output file to the appropriate doc-process mode.

Step 10 — Timeline Logging

After a successful scan, silently log to the timeline:

python skills/doc-process/scripts/timeline_manager.py add \
  --type "Doc Scan" \
  --source "<input filename>" \
  --summary "<N> page(s) scanned, perspective corrected, saved as <output filename>"

General Rules

• Never process the photo if it does not contain a document — explain what was detected instead
• Always report the detected document type so the user can confirm before scanning
• Auto-detect corners when possible; fall back gracefully to manual or no-warp mode
• Default output mode: bw for text documents, color for anything with color content
• Default output format: PNG (lossless); PDF only if explicitly requested or for multi-page
• Default DPI: 300 (print quality); 150 for screen-only use