Text-CAPTCHAs Classification using Deep Learning
Tejeshwani Singh, Amit Kumar, Paurav Goel · 2024
The categorization of text-based CAPTCHAs is essential for discerning between human users and automated bots on the internet. This procedure guarantees strong security, and this study uses deep learning methodologies for the classification of the text. This study uses open source text-CAPTCHAs dataset from Kaggle, and the dataset consists of more than 1000 images. This study investigates the use of SCNN and SMOTE techniques, in conjunction with ADAM, ADADELTA, and ADAGRAD optimization, to improve the precision and robustness of CAPTCHAs systems. In this study, the SMOTE technique is also used for image augmentation purposes, and image flips at 5-degree angle. This study proposed an SCNN model with ADAM optimizer for the CAPTCHAs text classification and reported the highest accuracy of 87.40%, and also experiment uses ADADELTA and ADAGRAD optimizer for the convergence purposes, but ADAM optimizers give the best performance, and for the overfitting condition, study uses dropout 0.2.