SecurePixels: An AI-Driven Chaotic Key Generation Framework for Secure Image Encryption in Sensitive Data Applications

Abstract

With the rapid proliferation of digital communication and Internet of Things (IoT) devices, the secure transmission of visual data has become critically important. Traditional chaotic image encryption schemes often rely on static chaotic maps, which can exhibit vulnerabilities when applied to images with varying statistical properties. In this paper, we propose a novel, intelligent image encryption architecture that utilizes Machine Learning (ML) to dynamically select the optimal chaotic diffusion map based on the intrinsic features of the plaintext image. The proposed algorithm extracts statistical features—including Shannon entropy, mean, standard deviation, skewness, and kurtosis—to drive a Random Forest classifier that predicts the most effective secondary chaotic map (e.g., Lorenz, Gauss, Henon, Chebyshev, or Logistic). The encryption process itself incorporates a multi-stage confusion-diffusion network. Initial confusion is achieved via bit-level permutation and rotation governed by a Tent map. To defend against autonomous channel attacks, the algorithm computes the Hamming distances between the Red, Green, and Blue (RGB) channels, encrypting these dependencies with a Logistic map mask. Finally, the dynamically selected chaotic map performs secure feedback diffusion. Furthermore, to ensure autonomous decryption without the need for supplementary key transmission, the architecture seamlessly embeds the identifier of the chosen map directly into the cipher image’s pixels using Least Significant Bit (LSB) steganography.