docscanify

n8n community nodes that scan and crop documents out of images (edge detection and perspective correction) and output them as images or PDFs

Package Information

Released: this month
Downloads: 0 weekly / 0 monthly
Latest Version: 1.2.0

Documentation

Logo

n8n-nodes-docscanify

This is an n8n community node that works like a "CamScanner": it takes a photo of a document, detects its edges, corrects the perspective and returns a PDF or an image.

The scanning is done by DocScanify, which lives in the same repository.

n8n is a fair-code licensed workflow automation platform.

Installation
Operations
Compatibility
Usage
Resources

Installation

Follow the installation guide in the n8n community nodes documentation. The published package's name is: @devantage/n8n-nodes-docscanify

The package has no install scripts, so it works with the n8n community nodes installer. Its native module, @napi-rs/canvas, ships prebuilt binaries for Linux (including Alpine), macOS and Windows.

Operations

These are the currently available resources and it's respective operations.

Document

  • Scan Document
Parameter Default Description
Input Binary Field data Binary field containing the image (JPG, PNG, WebP and other common formats)
Output Format pdf pdf (document inside a PDF page) or image (only the scanned image)
Output Binary Field data Field where the result is written
File Name (empty) Output file name. Empty uses the image name with the extension of the output format
Filter none none, grayscale or bw (document mode, adaptive threshold)
Reduce Noise false Smooths the noise, preserving the edges of the text
Sharpen off off, soft or strong
Rotation 0 Rotates the scanned document clockwise: 0, 90, 180 or 270 degrees
Page Size a4 PDF only. a4, letter or image (page sized to the image at 150 DPI)
Margin (Pt) 0 PDF only. Page margin
Max Dimension (Px) 3000 Downscales large photos before processing
JPEG Quality 85 Quality of the JPEG image (does not apply to bw, which uses PNG)
If No Document Detected error error or original (uses the whole photo)

The document is cropped, then enhanced (noise, sharpness and rotation, in this order), then filtered. With the image output format, the result is a JPEG, or a PNG when the bw filter is used. PDFs embed the image the same way.

The output JSON gets scan.detected, scan.corners (in original image pixels, even when the document is rotated), scan.width and scan.height.

With the pdf output format, it also gets scan.pdf, how the document was laid out on the page, in points (1 pt = 1/72 in), measured from the top left corner of the page:

{
  "unit": "pt",
  "page": { "width": 595.28, "height": 841.89, "orientation": "portrait" },
  "image": { "x": 56.69, "y": 120.4, "width": 481.9, "height": 601.1 },
  "pixelsPerPoint": 1.68
}

page already accounts for landscape pages, used when the document is wider than it is tall. image is where the document was drawn. Services that place fields on a PDF, such as e-signature platforms, ask for positions in points: a point found in the document image at px, py pixels is at image.x + px / pixelsPerPoint, image.y + py / pixelsPerPoint on the page. If a service measures y from the bottom of the page, use page.height minus that value.

Compatibility

The current version is compatible with n8n installations above v1.0.0.

This package replaces @devantage/n8n-nodes-scanify. It keeps the parameters and the output of its node, and adds the Reduce Noise, Sharpen and Rotation options. n8n identifies community nodes by their package, so workflows that use the old package need its node replaced with this one.

Usage

How detection works

Detection based only on the largest contour fails on textured backgrounds (granite, wood, fabric), because the document outline blends into the texture. So DocScanify:

  1. shrinks the image to about 800 px
  2. uses morphological closing and opening to erase the document's text and the background texture
  3. generates candidates from three masks (Otsu, inverted Otsu and Canny)
  4. accepts only convex quadrilaterals and scores each one by the fraction of its perimeter that lies on real edges
  5. picks the largest well-supported quadrilateral. If none qualifies, it falls back to the largest contour of the image

Notes

  • OpenCV.js is loaded once per process (about 1 s on the first run). Later runs reuse it.
  • Detection works best with the document on a contrasting background and all 4 edges visible.

Resources

Discussion