CAPTCHA VULNERABILITY ASSESSMENT USING CONVOLUTIONAL NEURAL NETWORKS
Gonzales-Quilca J.C.; Chalco-Huarachi E.A.; Vilca-Huayta O.A. · Revista de la Facultad de Ciencias (2026) · DOI: 10.15446/rev.fac.cienc.v15n2.117839
Resumen
The Completely Automated Public Turing Test to Tell Computers and Humans Apart (CAPTCHA) is a security mechanism designed to distinguish human users from automated bots and is widely employed to prevent unauthorized access. The objective of this study was to develop two models for recognizing text-based CAPTCHAs: Optical Character Recognition (OCR) and Convolutional Neural Networks (CNNs), which achieved accuracies of 37% and 99 %, respectively. A dataset comprising 10,000 CAPTCHA images generated using the Gregwar library was employed for training and evaluating the proposed models. The developed models proved effective in recognizing distorted and noisy text, while CNNs, in particular, demonstrated high robustness in text-based CAPTCHA recognition. Consequently, for systems that currently rely on the Gregwar library, it is recommended to replace it with alter-native CAPTCHA libraries that remain more resistant to automated attacks, such as image-based CAPTCHA schemes.
Instituciones: Universidad Nacional del Altiplano