Enhancing Alphanumeric Based CAPTCHA Solving using Deep Learning Techniques.

Abstract Today, captcha is a common method for preventing automated programs from accessing online resources like websites, emails, etc. Captcha is an acronym for Completely Automated Public Turing test to differentiate between computers and humans. It is a challenge-response test that determines whether the user is a person or a computer. There are various varieties of captcha techniques, including audio-based, text-based, and image-based, etc. Image-based captchas, in which the user is required to recognize and enter a combination of letters and numbers displayed in an image, are extensively used among these techniques. In this endeavor, we’ve employed deep learning techniques to improve the precision of alphanumeric-based captcha solving. We used a real-time dataset to make predictions with Convolutional Neural Network (CNN), Deep Neural Network (DNN), Combination of CNN and Recurrent Neural Network (RNN), and Neural Network with Nested Design model. TensorFlow, Keras, OCR, OpenCV, and Tkinter were used to create the model, and Tkinter is used for the image selection user interface. We used a dataset from Kaggle in which all the images are five-letter words that may contain digits, and noise (blur and a line) has been applied to the images. The images are PNGs measuring 200 by 50 pixels. We divided the dataset into train and test sets, calculated the accuracy, and provided an image captcha prediction. The goal of the project is to increase the accuracy of the model by comparing the outcomes of various deep learning techniques. This project aims to determine which deep learning technique provides the highest level of accuracy for solving alphanumeric-based captchas by comparing the accuracy of various deep learning techniques. This project’s outcomes can be utilized to improve the security of other captcha techniques.