In this section, we are going to work through creating CAPTCHAs through a generative adversarial network. We are going to generate images similar to those in the Street View House Numbers dataset (SVHN dataset). The idea is to use these GAN generated images as CAPTCHAs. Only when we have trained the GAN would they be easy to sample from a noise distribution. This would alleviate the need to create CAPTCHAs through a more complicated method. It would also provide some variety to the SVHN street numbers used in CAPTCHAs.
The SVHN is a real world dataset that is very much popular in the machine learning and deep-learning field for its use in the object recognition algorithm. As its name suggests, the dataset contains real images of house numbers obtained from Google Street View Images. The dataset can be downloaded from the following...
