Suliman a. Alsuhibany, Qassim University, Saudi Arabia

As automated optical character recognition (OCR) and deep learning-based solvers achieve near-human accuracy in breaking conventional CAPTCHAs, there is a critical need for security mechanisms that exploit the inherent limitations of machine perception. This paper proposes a novel methodological framework for "Calligraphic Obfuscation," a security-by-design approach that leverages the structural complexity and fluid entropy of traditional Arabic calligraphic styles. Unlike standard text-based challenges, our approach introduces a multi-phase generation pipeline that systematically maps linguistic strings into high-complexity visual domains. The methodology integrates a four-tier classification of calligraphic fonts—ranging from high-legibility styles like Naskh to high-entropy scripts such as Shakstah—and augments them with an adversarial layer utilizing Jacobian-based Saliency Map Attacks (JSMA). By formalizing the transition from cloud-centric generation to resource-efficient on-device architectures, this study provides a repeatable blueprint for developing robust, human-interactive proofs. The proposed framework offers a dual-benefit: significantly increasing the computational cost for adversarial machine learning models while maintaining a sustainable cognitive load for human users. This work lays the foundation for a new generation of linguistically-diverse and adversarially-hardened authentication challenges tailored for modern, resource-constrained mobile environments.

Calligraphic Obfuscation, CAPTCHA Security, Adversarial Machine Learning, Arabic Script Complexity, Human-Interactive Proofs, JSMA.