The Capsecure Project: Analyzing and Recognizing Captchas using a Custom-Based CNN Model

Abstract

To prevent attacks from programs or other computerized agents that try to mimic human intellect, CAPTCHAs are automated tests that try to discriminate between computers and people. The overarching goal of this study is to create a CAP-SECURE, a custom-based convolutional neural network (CNN) model, that can bypass CAPTCHA. Differentiating or informing websites about the CAPTCHAs' shortcomings and vulnerabilities is the goal of the suggested approach. Compared to other CNN architectures, such as VGG-16 and ALEX-net, the CAP-SECURE model—which is based on a sequential CNN model—performs better. In theory, the model could decipher and investigate both numerical and alphabetical CAPTCHAs. In order to train our model effectively, we have created a dataset consisting of 2,00000 CAPTCHAs. To address the present problems and give solutions for CAPTCHAs, we examine a CNN-based deep neural network model in this presentation. For the alpha-numerical test dataset, the network's cracking accuracy is shown to be 94.67%. The suggested custom-based model outperforms more conventional deep learning approaches in terms of both recognition rate and resilience.