Sanapala Joshika

https://doi.org/10.64823/ijter.2503015

This paper presents an intelligent system for solving CAPTCHA challenges using Artificial Intelligence (AI) integrated with explainable frameworks. CAPTCHAs, designed to differentiate humans from bots, often pose accessibility and usability issues. To address this, we developed a deep learning model capable of accurately recognizing and solving both alphanumeric and math-based text CAPTCHAs. The model utilizes Convolutional Neural Networks (CNNs) for image-based text recognition, trained on synthetically generated CAPTCHA datasets. To enhance transparency and trust in AI predictions, the system incorporates LIME (Local Interpretable Model-agnostic Explanations), which visually explains each character prediction by highlighting important regions of the CAPTCHA image. This interpretability aids developers in validating the model's decisions and ensures robustness against adversarial inputs. The system aims to balance accuracy, security, and explainability, making it suitable for real-world applications where both user experience and AI accountability are critical.