Cryptanalysis and Improvement of an Image Cryptosystem via Chaotic Synchronization and Artificial Neural Networks in Modified OTP Scheme

Abstract

Chaos-based image ciphers may exhibit favorable statistical metrics while retaining a recoverable algebraic structure. We analyze the modified one-time-pad image cryptosystem of Stoycheva et al., which combines static, Chua-system-derived, and neural masks with channel-wise shuffling and XOR. For observations sharing a fixed or reproduced induced state, we prove that each color channel is equivalent to a plaintext-independent permutation followed by an XOR keystream. The trained neural component supplies a mask but does not alter this external structure. Established base-256 position labels recover the complete equivalent mapping with four chosen-plaintext queries for a 512-by-512 image; known-plaintext, chosen-ciphertext, differential, and black-box analyses are evaluated under explicitly stated access conditions. Under the same state, a one-byte plaintext difference changes exactly one ciphertext byte. We also specify a redesign that renews the three key components per image and couples cross-channel routing and masking to preceding ciphertext bytes. Its inverse is proved correct. Across 15 image–function trials, decryption was byte exact. Advancing the image counter changed 99.603907% of ciphertext bytes for the same image, and none of eight tested input pairs satisfied the fixed permutation–XOR affine identity. These results identify the state condition required by the recovery procedures and characterize the tested behavior of the redesigned image transformation.