This project creates a convolutional neural network based architecture to determine the three characters present given a CAPTCHA image. Note that a blog post about this project has been written here.
CAPTCHA Solver - End-to-end CNN Approach.ipynb
is a Jupyter notebook that can be run which contains more technical detail and implementation of this model.
Training dataset accuracy: 98.95% Validation dataset accuracy: 99.74%
Thank you to Geetika Garg's for the model architecture used in this project. In addition, thank you to fateleak for providing a decent size CAPTCHA image dataset that can be found here.