Implementasi Model Convolutional Neural Network (CNN) dan Resnet-50 dalam Identifikasi Captcha Teks Terdistorsi

Abstract

Text-based CAPTCHA is a commonly used security system to distinguish between human users and bots by displaying distorted text. This study explores the recognition of distorted text CAPTCHAs using deep learning approaches, focusing on implementing the CNN and ResNet-50 models. Through this implementation, the research demonstrates that recognition accuracy improves significantly, revealing vulnerabilities in text-based CAPTCHA systems. The implemented model achieved 96.78% accuracy on 4-character CAPTCHAs, 94.86% accuracy on 5-character CAPTCHAs, and 93.11% accuracy on 6-character CAPTCHAs. These findings highlight the importance of developing stronger and more innovative CAPTCHA systems to maintain the security of online platforms. Future studies could benefit from these results by exploring other CAPTCHA variants and more advanced deep learning techniques, as well as balancing security and user experience