Synthetic Text Images Generator

Generate image triplets [anchor, positive, negative] with random text to train a triplet loss function for different purposes:

• Text Recognition
• Text Similarity from images

note that text in anchor & positive is same while it's different in the negative.

Requirements

• pip install opencv-python
• image augmentor https://github.com/aleju/imgaug, pip install imgaug

things to do/add:

I am planning to add:

• generate text from file. (currently only random text)
• more image augmentation options
• support other languages other than english (currently opencv support only English, will use different library to write text on image)
• exposue all thresholds in a config file with a documentation for each of them