RIFT-PWLCM: A Reproducible Fixed-Point Image-Encryption Construction with Domain-Separated Bidirectional Feedback and Authenticated Output

Abstract

Many chaos-based image ciphers obtain excellent NPCR, UACI and entropy values while retaining floating-point maps, genetic tuning, DNA-labelled operations, global sorting, or unauthenticated ciphertext. This work introduces RIFT-PWLCM (Reversible Integer Feedback Transformation with PWLCM-derived material), a reproducible research construction that removes those elements. A 256-bit key and 128-bit nonce are domain-separated with SHAKE-256 into public-parameter Q0.32 integer PWLCM-derived material for four roles per round (PRE, FWD, BWD and POST). Six reversible rounds combine pre/post whitening with AES S-box forward and backward byte-feedback recurrences. Ciphertext and associated data are protected by an encrypt-then-MAC HMAC-SHA-256 tag truncated to 128 bits. All reported values were regenerated from the stated implementation. Across seven 256×256 grayscale inputs and one 256×256 RGB input, ciphertext entropy is 7.996996–7.999106 bits/byte and the largest absolute adjacent correlation is 0.00927. Over 105 single-bit plaintext trials, NPCR is 99.6080±0.0263% and UACI is 33.4758±0.0885%. Exhaustive single-bit flips of the 256-bit key and 128-bit nonce change 49.9926% and 49.9970% of ciphertext bits on average. Generator autocorrelations through lag 32 remain below 0.00533 in magnitude; bounded differential, linear, NIST SP 800-22, weak-key, nonce-reuse, and authentication screens are also reported. All 10,000 ciphertext-tamper trials were rejected and no false acceptance occurred in 100,000 random-tag trials. The Python/Numba reference implementation reaches roughly 0.42–0.55 MB/s for medium-to-large buffers. These experiments establish reproducibility and feasibility, not a formal security level; RIFT-PWLCM is not proposed as a replacement for standardized AEAD such as Ascon.