A program to automatically crack the challenges of Google's reCAPTCHA. Made as part of my dissertation for my BSc in Computer Science at the University of Sheffield, with the help of my supervisor Dr Richard Clayton and my second marker Dr Tony Simons. The code of this project was written by me, but with guidance from them on the direction of my project, and feedback on my report.
- If you have Pipenv installed, there is an included Pipfile, so you can create a Python3 environment using
pipenv --three, and then you can runpipenv install - If you don't have Pipenv, then consult the Pipfile and Pipfile.lock for which packages to install using Pip (also consider installing Pipenv it's great: https://github.com/kennethreitz/pipenv/)
- You can either use the existing labels, categories and weights files or provide your own in the same format. If you do the latter, make sure to change config.py. If you wish to train a new network using this code, you will also need to change config.py and supply a suitable dataset.
- In order for Splinter to work, you will need geckodriver, which is the web driver behind Firefox, the default browser for Selenium.
- There is a possibility for issues on Windows due to the lack of jpeg support in Pillow on Windows without libjpeg, which is difficult to install on Windows. This page might help: https://stackoverflow.com/questions/20672530/installing-pillow-on-windows-fails-to-find-libjpeg
- Once you have installed dependencies, you can run the CAPTCHA cracker by changing to the captcha-cracker directory and using
python3 captcha_input.py(Assuming your python 3 install is ran by runningpython3) - To view CAPTCHAs previously seen by the CAPTCHA cracker, change directory to captcha-predictions-viewer and run
python3 main.py - To retrain the network, instantiate a NeuralNetwork with a learning and decay rate, or uncomment the example at the bottom of nn.py and run nn.py