(v 1.0) | AI, Web Security

unCAPTCHA is a simple ML algorithm used to read and write distorted images from FakeCAPTCHA.com.

As of now, the recognition rate of said ML algorithm is about a 97% recognition rate. Being much more faster and considerably better recognition by using data from FakeCAPTCHA.com.

Though, the ML algorithm can only recognize digits and not other characters.

Final Report

Prerequisites:

• Python 3.x
  • NumPy 1.19.x
  • cv2 4.5.x
  • Matplotlib 3.4.x
  • Tensorflow 2.4.x with Keras
• Anaconda 4.9

Activating Conda Virtual Environment

Before cloning this repository, we advise you to create a separate virtual environment for the ML algorithm to run properly, of course, if you know what you are doing, skip to Clone the Repo and Running the Program. This can be done using Anaconda.

$ conda create -n tf tensorflow
$ conda activate tf

This is will help not create any internal issues with non-supercomputers. Now install all the listed pip3 packages needed to run the algorithm, assuming NumPy and TensorFlow are already installed within the virtual environment.

(tf) $ pip3 install opencv-python
(tf) $ pip3 install matplotlib

Cloning the Repo and Running the Program

$ git clone https://github.com/tardiobscurus/unCAPTCHA.git
...
$ cd unCAPTCHA

Then run uncaptcha.py with a generated CAPTCHA, in this case, we will add fc001.jpg (must be either jpg or png)

$ python3 segment.py fc001.jpg

There are a few errors with uncaptcha.py, is a bit buggy with matplotlib.