Hyperchaotic Image Encryption Scheme Based on Deep Neural Models and Chaos-Driven Key Streams
No Thumbnail Available
Date
2026
Journal Title
Journal ISSN
Volume Title
Publisher
Library Information Services, COMSATS University Islamabad, Lahore Campus
Abstract
With the rapid growth of digital communication, cloud storage, telemedicine, and multime-
dia transmission, the secure protection of digital images has become an important research
problem. Conventional encryption methods are not always suitable for image data because
images contain high redundancy, strong correlation among neighboring pixels, and large
data size. To address these challenges, this thesis proposes secure and efficient color image
encryption schemes based on chaotic systems, fractional-order hyperchaotic dynamics, and
deep neural models.
The first proposed scheme uses a two-dimensional Sine Logistic Adjusted Map (2D-
SLAM) to generate chaotic key streams for color image encryption. The method combines
plaintext-dependent key generation, row and column scrambling, Josephus traversing, cat-
map-based permutation, and forward-reverse diffusion to achieve strong confusion and dif-
fusion. The second proposed scheme integrates a fractional-order memristive hyperchaotic
system with a lightweight convolutional neural network. In this framework, neural latent
features are used to construct image-adaptive dynamic S-boxes, while hyperchaotic se-
quences are employed for key scheduling, permutation, substitution, and diffusion. The use
of a public nonce and masked latent header further strengthens resistance against keystream
reuse and plaintext-related attacks.
Extensive experimental analysis is performed on standard color images of different
sizes. The results show that the proposed schemes produce cipher images with uniform
histograms, entropy values close to the ideal value of 8, very low adjacent-pixel correla-
tion, large key spaces, and strong key sensitivity. Differential attack analysis gives NPCR
values close to 99.6% and UACI values close to 33.4%, demonstrating strong sensitivity to
ix
small changes in the plaintext. Furthermore, the decrypted images are accurately recovered
under normal decryption, while robustness tests against noise and data-loss attacks show
the stability of the proposed methods.
Overall, the results confirm that the proposed encryption schemes provide high secu-
rity, strong randomness, and reliable reconstruction performance. Therefore, the proposed
approaches are suitable for secure transmission and storage of color images in modern
multimedia communication systems.
Description
Keywords
Department of Mathematics, Mathematics, FA24, Image Encryption, Hyperchaotic Systems, Deep Neural Networks, Chaos-Based Cryptography, Dr. M. Faisal Nadeem
