扫描仪

When to Use

Use when the user wants to scan a document from a photo, crop the page automatically, fix perspective, or make an image look like it came from a flatbed scanner.

Load this skill when the request sounds like:

• - "scan this document"
• "crop the page edges"
• "make this photo look scanned"
• "deskew this paper"
• "clean up this receipt or sheet"

Default Engine

Default to jscanify.

Why this is the default:

• - Open source and MIT licensed
• Mature enough to trust for general use
• Works in browser, CDN, npm, and Node-oriented workflows
• Uses OpenCV.js without forcing a native OpenCV install
• Lower friction than mobile-native or commercial SDKs

Do not default to commercial SDKs unless the user explicitly asks for a production mobile SDK or already uses one in their app.

Quick Workflow

1. 1. Start with jscanify, not custom image math.
2. Detect the page boundary first.
3. Extract with perspective correction.
4. Review the output visually.
5. If edges are wrong, stop and retry with a cleaner input or a manual-corner flow instead of stacking random filters.

Basic Commands

1. Project install

CODEBLOCK0

2. Zero-global-install local preview

Use CDN scripts in a tiny HTML page and serve it locally:

CODEBLOCK1

Load:

CODEBLOCK2

3. Core extraction

CODEBLOCK3

4. Edge preview before crop

CODEBLOCK4

Core Rules

1. Browser-first wins

• - Prefer the browser or browser-like path first because jscanify is strongest there.
• Prefer CDN or local project dependency over native OpenCV builds.
• If the user only needs one scan, choose the lowest-friction path.

2. Detect before you "enhance"

• - First run boundary detection on the original image.
• Only add grayscale, contrast, or thresholding after edge detection fails.
• Do not overprocess before trying the default extraction path.

3. Expect flat, visible borders

• - Best results come from one document on a contrasting background.
• Top-down or near-top-down photos are safer than steep angles.
• Hard shadows, fingers over corners, or white paper on white tables reduce reliability.

4. Preserve the original

• - Never overwrite the only copy of the input image.
• Save the corrected export separately.
• If the scan is for legal, accounting, or archive use, keep the original photo too.

5. Manual correction beats fake certainty

• - If auto-detection is clearly wrong, do not pretend it succeeded.
• Ask for a better photo or switch to a manual-corner workflow.
• Wrong corners produce worse output than no crop.

Common Traps

• - Busy backgrounds can confuse edge detection.
• Low contrast document-on-table shots fail more often than users expect.
• Glare can hide one full edge and make the crop collapse.
• Very curved paper is not a normal document-scan case.
• Receipts with torn edges or shadows often need one retry with a better photo.
• OCR and document scanning are different jobs; scan first, OCR second.

Scope

This skill ONLY:

• - Chooses and applies a document-scanning workflow
• Prioritizes jscanify by default
• Produces cropped, perspective-corrected scan-style images

This skill NEVER:

• - Claims OCR accuracy improvements by itself
• Recommends paid SDKs by default
• Reimplements document detection from scratch unless the user explicitly wants that

External Endpoints

Related Skills

• - documents - General document handling after the scan is cleaned.
• INLINECODE9 - Extra image cleanup when the user needs post-processing beyond edge detection.
• INLINECODE10 - File organization, renaming, and export handling after scans are generated.

Feedback

• - If useful: INLINECODE11
• Stay updated: INLINECODE12