Enhanced cast-128 with adaptive s-box optimization via neural networks for image protection (2509.07606v1)

Abstract: An improved CAST-128 encryption algorithm, which is done by implementing chaos-based adaptive S-box generation using Logistic sine Map (LSM), has been provided in this paper because of the increasing requirements of efficient and smart image encryption mechanisms. The study aims to address the drawbacks of static S-box models commonly used in traditional cryptographic systems, which are susceptible to linear and differential attacks. In the proposed scheme, the dynamic, non-linear, invertible, and highly cryptographic strength S-boxes are generated through a hybrid chaotic system that may have high non-linearity, strong and rigorous avalanche characteristics, and low differential uniformity. The process here is that the LSM is used to produce S-boxes having key-dependent parameters that are stuffed into the CAST-128 structure to encrypt the image in a block-wise manner. The performance of the encryption is assessed utilizing a set of standard grayscale images. The metrics that are used to evaluate the security are entropy, NPCR, UACI, PSNR, and histogram analysis. Outcomes indicate that randomness, resistance to statistical attacks, and country of encryption are significantly improved compared to the original CAST-128. The study is theoretically and practically relevant since it presents a lightweight S-box generation approach driven by chaos, which can increase the level of robustness of the image encryptions without enlisting machine learning. The system may be applied to secure communications, surveillance systems, and medical image protection on a real-time basis.