captchaSolver

Solving CAPTCHAs using 2 different Neural Networks (Pre-Trained and Self-Trained). Captcha Solver is an exercise for CICE´s Ethical Hacking Masters Degree - EC-Council & CompTIA. MSI.

Idea

Explanation of different methods using Machine Learning to solve some CAPTCHA security measures.

Machine Learning Solution

The first exercise can be found in HackingCAPTCHAS_MachineLearning.ipynb We try to hack one of Wordpress most used CAPTCHA Plugins called Really Simple CAPTCHA. Analyzing the source code of the plugin we find a method to generate CAPTCHA images and their solution. We use this information to train a Deep Learning Convolutional Neural Network so it can solve any other random CAPTCHA produced by this plugin. Using Keras and TensorFlow we use a 10.000 training data set combined with OpenCV to detect individual letters. After training, the neural network proves being able to solve this sort of CAPTCHA images with a 99.89% accuracy.

Machine Learning Pre-Trained Solution

In this second example, we use Web Scrapping to try and complete a web form protected with a CAPTCHA image a human must solve. We use a LSTM Neural Network, combined with Google´s Tesseract OCR Engine to solve the CAPTCHA and automate the form fill process. The network is pre-trained with 400.000 individual characters in over 100 languages. In this other example, the randomness of the CAPTCHA generator is much bigger than the one in the previous example. Therefore, it must be more secure. Nevertheless, using our Neural Network, we prove it solves the CAPTCHA 92% of the time.

Project Requirements

To fully test the project, you might need to install some required packages. Please start by downloading Anaconda using the following link: Download Anaconda. Once installed, clone this repository and access the downloaded folder through a terminal window. Next, you need to install the required packages:

$ conda env create -f pmateoCAPTCHA.yml

Once all the packages have been installed, activate the environment:

$ source activate pmateoCAPTCHAS

Once finished, you can check the Jupyter Notebooks I have created

 (pmateoCAPTCHAS)$ jupyter notebook

There are 2 Jupyter Notebook Files to check:

• HackingCAPTCHAS_MachineLearning.ipynb
• HackingCAPTCHAsPreTrained.ipynb