Font from Image

Supervised Learning - Learn Font names from text in an image

This is a cleaned up version of experimentations I did in 2018 after my Deep Learning Specialization with Andrew Ng in 2018.

Having played a lot with fonts and drawing texts on an image for my Mix on Pix app, I wanted to see if it was possible to use Deep Learning to learn what is the font used in an image.

So I created a 3 steps process:

1. Generate images that contain a text from a variety of fonts.
2. Do Extract-Transform-Load (ETL) preprocesing to have the data ready for Learning.
3. Train a model for classification to identify the font in an image.

---

After classification, I created an interesting Saliency Map

---

Directories

• data/font_data: The Dataset of 11,760 images. It is composed of:
  • 1 directory per Font. For each of the 35 fonts, we have:
    • 86 images with 1 character per image. 50px x 50px.
    • 250 images with 3 characters per image. 150px x 50px.
• ImagesGeneration: macOS application. This application generates images with text. Configurable.
• notebooks: Jupyter Notebooks for:
  • ETL to generate 29,260 50x50x1 images and labels.
    • 35 * (86 + (3 * 250)) = 29,260
  • Performs the actual training to identify the font in the text of an image.

---

Communication

• If you want to contribute, submit a pull request.
• If you found a bug, have suggestions or need help, please, open an issue.

---

Licenses

• The data: Font from Image (FFI) Dataset © 2022 by Francois Robert is licensed under CC BY 4.0. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
• The code: The code in the notebooks and the macOS code to generate the images are licensed under the MIT license. See here and here.

---

by Francois Robert