This code was tested on the python 3.
Using the really sample captcha dataset. Dataset is easily available at the website "htttp://wordpress.org". this dataset has 800,000+ active installations. Source code to the wordpress login is already available. Image are in png format
Decoding the captcha code using CNN with two convolutional layers and two fully connected layers.
• Designed a Deep Learning CNN architecture network for computer vision that breaks the text-based CAPTCHA’s by using TensorFlow an open-source library for machine learning.
• Used pooling layers to make the model more dynamic and robust.
• The designed and developed model trains with 20% of total feasible data and cracks the current text-based captcha’s with a high accuracy rate of 97.8%, altering the security concerns to online transactions.