Chen Wei, Wang Yichuan, Shi Cheng, Sheng Guanglei, Li Mengyang, Liu Yu, Hei Xinhong
School of Computer Science and Engineering, Xi'an University of Technology, Xi'an 710048, China.
School of Computer Science and Engineering, Xi'an University of Technology, Xi'an 710048, China; Shaanxi Key Laboratory for Network Computing and Security Technology, Xi'an 710048, China.
Neural Netw. 2025 Nov;191:107799. doi: 10.1016/j.neunet.2025.107799. Epub 2025 Jul 4.
As digital images are extensively applied across diverse domains, the demand for visually secure image encryption technology has surged remarkably. However, existing schemes generally suffer from insufficient encryption security and low-quality decrypted images. Therefore, this paper proposes a flexible scheme that integrates meta-learning, a chaotic system, traditional deep learning, and the LSB-2 correction embedding method. The core of this scheme lies in the design of a meta-learning compression reconstruction network with dynamic auxiliary input, which enables high-quality compression of a plain image. Then, a novel chaotic system, IS-DP, is constructed to encrypt the compressed image into a noise-like secret image by combining 2D-IS chaotic system with a traditional deep learning network. Finally, a lossless embedding method with LSB-2 correction is employed to embed the secret image into a carrier image, resulting in a visually secure cipher image. This scheme fully validates the great potential and feasibility of deep learning methods in encryption and compression. Moreover, the flexibility endowed by the meta-learning mechanism allows users to adjust the inner-loop iteration number according to practical needs, balancing running time and decrypted image quality, thus demonstrating broad application prospects.
随着数字图像在各个领域的广泛应用，对视觉安全图像加密技术的需求急剧增加。然而，现有方案普遍存在加密安全性不足和解密图像质量低的问题。因此，本文提出了一种灵活的方案，该方案集成了元学习、混沌系统、传统深度学习和LSB - 2校正嵌入方法。该方案的核心在于设计一个具有动态辅助输入的元学习压缩重建网络，该网络能够对明文图像进行高质量压缩。然后，构建了一种新型混沌系统IS - DP，通过将二维IS混沌系统与传统深度学习网络相结合，将压缩后的图像加密成类似噪声的秘密图像。最后，采用具有LSB - 2校正的无损嵌入方法将秘密图像嵌入载体图像，得到视觉安全的密文图像。该方案充分验证了深度学习方法在加密和压缩方面的巨大潜力和可行性。此外，元学习机制赋予的灵活性允许用户根据实际需求调整内环迭代次数，平衡运行时间和解密图像质量，从而展现出广阔的应用前景。
