Signature Extractor

Extract clean, transparent-background signature ink from photos. Use when: user provides a photo of a handwritten signature and asks to remove the background, make signature transparent, upscale resolution, deepen/darken ink, or remove red stamps/seals. NOT for: general photo background removal (use remove.bg or similar), color signature ink extraction, or document scanning/OCR tasks.

Extract handwritten signature ink from photographs or scanned documents, producing a clean transparent-background PNG. Removes white/light backgrounds, red stamps/seals, and colored paper textures while preserving black ink strokes.

When to Run

• User provides an image and asks to "remove background from signature", "make signature transparent", "extract signature", "remove stamp/印章"
• User asks to "darken signature ink", "make ink bolder", "填满笔迹", "加深签名"
• User asks to "upscale signature", "make signature clearer", "提高签名清晰度"
• User says "去掉底色", "去除背景", "签名透明", "提取签名", "去除印章"

Workflow

1. Identify the user's desired mode from their request:

   Request keywords	Mode	What it does
   "填满", "加深", "实心", "加粗", "solid", default	solid	3x upscale + sharpen + pure black opaque ink
   "高清", "放大", "清晰", "upscale", "hd"	hd	3x upscale + sharpen + smooth alpha gradient
   "去掉底色", "提取", "简单去掉", "quick", "extract"	extract	Original size extraction with alpha gradient

2. Run the extraction script:

   python3 {baseDir}/scripts/extract_signature.py INPUT_IMAGE OUTPUT.png --mode MODE [--scale N]
   

   • Default scale is 3 (for solid and hd modes). Adjust --scale if user specifies a different multiple.
   • Output is always RGBA PNG format.

3. After completion, report: output dimensions, ink pixel count, and share the result with the user.

How It Works

The script isolates black ink using a three-channel color filter:

1. Brightness filter — pixels darker than a threshold are ink candidates
2. Red channel filter — R minus G must be less than 40, which excludes red stamps/seals
3. Saturation filter — low saturation ensures only true black/grey ink is captured, not colored marks

For solid and hd modes: LANCZOS upscaling → UnsharpMask sharpening → ink extraction → Alpha channel smoothing.

Prerequisites

pip install Pillow numpy