This is a captcha generator created for NUS CS4243 mini project. It uses Wasserstein Generative Adversarial Network (WGAN) and generates individual characters. The dataset that we used to train is not included in this directory. Below is a sample output after training the model for 1000 epochs with main.py.

After generating the individual characters, we used a OCR tool to filter out ambiguous characters with check_image_validity.py. If the output is still not satisfying, we can manually label each images generated with manual_label.py. Lastly, we ensemble the characters into captchas by adding colours, distracted blacklines and changing their proportions in process_captcha.py.

We have also included our trained model in model folder.