Beating Google Captcha with Artificial Intelligence

John T Haag, Zach Kovalenko, Thomas Shear, Dr. Ning Yu, Department of Computer Science, SUNY College At Brockport, 350 New Campus Dr, Brockport, NY 14420, United States

The Turing test was named after the godfather of computer science Allan Turing. This important concept in computer science begs the question, whether or not a computer can tell the difference between the intelligence of a human and another computer. The Google Captcha (Completely Automated Public Turing test to tell Computers and Humans Apart) is an example of a Turing test. The Google Captcha was created to avoid spam, and robots filling in important data fields within websites. Our project is aiming to break the Google Captcha using modern Artificial Intelligence techniques. This will allow continued use of robots on websites that previously challenged the use of robots. The methodology used for our project comes in the form of object detection models. These models use Artificial Intelligence techniques such as Neural Networks to find a Google Captcha anywhere on a computer screen and output a percentage of likely hood it is a Captcha along with a label to the desired item. Our goal is to train two object detection models to find a Google captcha and build a computer bot to click on the Captcha without any human interaction. Thus, solving the Captcha “I am not a Robot” without images. Some research shows that a new type of object detection model may beat a classic TensorFlow model. Thus, one model will be created with TensorFlow and Deep learning, and the other with Pytorch and a YOLO (You Only Look Once) version 5 (v5) Neural Network. Our expectation is that the YOLO v5 model will beat the TensorFlow model in many benchmark measurements including a loss of less than 0.05, a high accuracy (more than 90%), and increase the speed at which testing the model occurs with a Nvidia GPU (Graphical Processing Unit).

Additional Abstract Information

Presenters: John Haag, Zach Kovalenko, Thomas Shear

Institution: State University of New York- Brockport

Type: Poster

Subject: Computer Science

Status: Approved

Time and Location

Session: Poster 5
Date/Time: Tue 12:30pm-1:30pm
Session Number: 4008