Handwriting Recognition Data Generator

Handwriting Recognition Data Generator is a utility that generates images that attempt to look like handwritten documents. The user can specify images to use as background and a source of glyphs images that will be stamped onto the page.

Getting started

pip install -r requirements.txt
python main.py -h

• Glyphs are expected to be located in folders named after their decimal representation (e.g., A => 65, a => 97)
• Glyphs for the same font are expected to have the same name in each folder (e.g., 65/00001.png, 97/00001.png)

Limitations

• Only a subset of Latin/English glyphs (! (ASCII 33) to ~ (ASCII 126))
• Only a single region (where glyphs are stamped)
• Glyph images are expected to be jpg or png

Configuration

backgrounds:
  -
    file: /path/to/file
    regions:
      [[100, 100], [500, 500]]
    line_height: 50
    dilate: 3
    blur: 0

• file: location of the background image on disk
• regions: array of arrays defining the regions where glyphs may be stamped, in the form (top left x, top left y, bottom right x, bottom right y)
• line_height: the height (in pixels) of the line in the background image
• dilate: amount of morphological dilate to apply to glyphs
• blur: amount of blur to apply to glyphs

License

The code is licensed under the MIT license. See LICENSE.