feedscanpp

A robust document and receipt scanner optimized for ADF and thermal paper.

Tested with Epson ES-C200 scanner using scanimage set to color mode.

Features

• Automatic Skew Correction: FFT-based skew detection to straighten tilted documents
• Smart Trimming: Removes white borders automatically
• Tint Removal: Eliminates yellowing and discoloration from aged thermal paper
• Intelligent Color Detection: Distinguishes between Color, Grayscale, and Black & White documents
• Adaptive Enhancement: Optimizes image quality based on detected content type

Installation

Using pip

pip install feedscanpp

Using uvx (no installation required)

uvx --from git+https://github.com/chazeon/feedscanpp.git feedscanpp scan-*.png -o output-{index}.png

From source

git clone https://github.com/chazeon/feedscanpp.git
cd feedscanpp
pip install -e .

Usage

feedscanpp scan-*.png -o processed-{index}.png

The output path supports placeholders:

• {index}: Sequential index (0, 1, 2, ...)
• {name}: Original filename with extension
• {stem}: Original filename without extension

Options

You can disable individual processing steps:

--no-rotate    Disable rotation correction
--no-trim      Disable trimming
--no-tint      Disable tint removal
--no-enhance   Disable enhancement

Example:

# Only apply rotation and trimming
feedscanpp scan.png -o output.png --no-tint --no-enhance

How It Works

The processing pipeline consists of four stages:

1. Rotation Correction

   • Detects skew angle using FFT magnitude spectrum analysis
   • Rotates image to align text horizontally (±5 degrees)

2. Trimming

   • Finds content bounding box by detecting non-white pixels
   • Removes excess white borders

3. Tint Removal

   • Estimates background color using morphological dilation
   • Normalizes brightness to eliminate yellowing
   • Applies CLAHE for local contrast enhancement

4. Analysis & Enhancement

   • Detects document type: Color, Grayscale, or Black & White
   • Applies Otsu's thresholding for B&W documents
   • Preserves color/grayscale when appropriate

License

MIT