Scholarlink

<p>In an era of increasing cyber threats, safeguarding sensitive data such as financial transactions and healthcare records demands innovative encryption strategies. This paper introduces CovertGAN-an AI-driven cryptographic framework that integrates deep learning, generative adversarial networks (GANs), and diffusion-based models to create a secure and adaptive system for data protection. The proposed approach merges biological classification with encryption, where bird species classification acts as a contextual trigger to enhance entropy and randomness in key generation. A GAN-based encryptor-decryptor architecture, trained in an adversarial setting, ensures robustness against eavesdropping attacks, while Stable Diffusion models transform encrypted data into visually meaningful cipher images. Experiments conducted using the Kaggle 400 Birds and Stanford Dogs datasets demonstrate that the system effectively resists adversarial inference, achieving optimal trade-offs between security and computational efficiency. This work presents a step toward intelligent, adaptive, and visually coherent cryptographic systems capable of evolving alongside modern attack vectors.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CovertGAN: intelligent cryptography using cross-domain adversarial generative diffusion

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Sridhar Iyer,
  • Narendra Shekokar

摘要

In an era of increasing cyber threats, safeguarding sensitive data such as financial transactions and healthcare records demands innovative encryption strategies. This paper introduces CovertGAN-an AI-driven cryptographic framework that integrates deep learning, generative adversarial networks (GANs), and diffusion-based models to create a secure and adaptive system for data protection. The proposed approach merges biological classification with encryption, where bird species classification acts as a contextual trigger to enhance entropy and randomness in key generation. A GAN-based encryptor-decryptor architecture, trained in an adversarial setting, ensures robustness against eavesdropping attacks, while Stable Diffusion models transform encrypted data into visually meaningful cipher images. Experiments conducted using the Kaggle 400 Birds and Stanford Dogs datasets demonstrate that the system effectively resists adversarial inference, achieving optimal trade-offs between security and computational efficiency. This work presents a step toward intelligent, adaptive, and visually coherent cryptographic systems capable of evolving alongside modern attack vectors.