With the recent success of convolutional neural networks (CNNs) in computer vision tasks, breaking basic CAPTCHAs in a few minutes is a relatively easy task. Consequently, CAPTCHAs need to be much more evolved than they have in the past. In the first part of this chapter, we are going to expose the vulnerability of basic CAPTCHAs being automatically detected using bots with a deep-learning framework. We are going to follow this up by exploiting GAN to create CAPTCHAs that are harder for bots to detect.
Breaking CAPTCHAs with deep learning
Generating basic CAPTCHAs
CAPTCHAs can be generated using the Claptcha package in Python. We use this to generate CAPTCHA images of four characters consisting of numbers and text. Consequently...