A research team has demonstrated an image encryption scheme that combines non-uniform one-dimensional cellular automata, dual chaotic maps, and dynamically generated substitution boxes to secure visual data directly on resource-constrained Artificial Intelligence of Things edge hardware. Modern networked systems increasingly rely on edge devices such as smart surveillance cameras, industrial monitors, and wearable sensors to collect continuous streams of visual information. Because conventional cryptographic systems like the Advanced Encryption Standard and Rivest–Shamir–Adleman require computational and memory allocations that exceed the operational limits of embedded edge processors, transmitting unencrypted or poorly protected imagery exposes sensitive information to statistical interception and differential cryptanalysis. The newly reported framework establishes a streamlined pipeline that achieves thorough confusion and diffusion across transmitted image matrices while maintaining linear time and space complexity.
To overcome the architectural limitations of low-power edge units, the researchers structured their encryption pipeline around lightweight mathematical transformations that avoid complex algebraic overhead. The encryption framework processes incoming data by first acquiring two-dimensional image matrices and subsequently executing four interconnected stages: key sequence generation, position-dependent pixel permutation, non-linear substitution, and chained XOR-based diffusion. Rather than relying on static transformation components that attackers can model over repeated observations, every stage of the proposed workflow operates dynamically. Control parameters and seed states govern how pixel values transition across the spatial grid, ensuring that subtle variations in plain input imagery yield entirely uncorrelated ciphertext without demanding excessive memory storage or iterative processing loops.
The chaotic key generation mechanism employs two independent dynamical systems running in parallel: a one-dimensional Logistic map and a two-dimensional Tinkerbell map. While a standard Logistic map provides fast execution and high sensitivity to starting conditions, its single-parameter formulation exhibits a restricted key space when implemented alone. By coupling it with the Tinkerbell map, which generates paired state variables via quadratic transformations, the architecture produces three distinct entropy streams per iteration. The continuous state outputs are scaled by a factor of 10 to the power of 14, floored, and converted into eight-bit integer representations modulo 256. A bitwise exclusive-OR operation then combines these streams into a single hybrid chaotic key sequence, significantly expanding the effective key space while increasing resistance against statistical analysis.
Spatial permutation within the framework is executed through a hybrid one-dimensional cellular automaton operating under a non-uniform rule vector specified as 204, 51, 51, 51, 195, 60, 60, and 51. Unlike uniform cellular automata that apply identical local transition logic across an entire lattice, the non-uniform arrangement assigns distinct elementary transition rules to different cell positions. Rules 51 and 204 provide complementary transition characteristics, whereas rules 60 and 195 introduce pronounced non-linear neighborhood interactions across three-cell spans. By breaking the predictable geometric dependencies and repetitive lattice cycles common to single-rule structures, this position-dependent permutation thoroughly decorrelates adjacent pixel relationships across both horizontal and vertical dimensions before data enters the substitution stage.
Following spatial scrambling, the framework introduces non-linear confusion using a dynamically constructed substitution box derived directly from the chaotic sequence. Generating a dynamic 15-by-17 matrix containing 256 hexadecimal entries eliminates vulnerabilities tied to static lookup architectures commonly used in traditional cryptographic engines. The chaotic stream is scaled to the discrete range between 0 and 255, filtered to discard duplicated numerical entries, and arranged into a strictly bijective mapping. Because the substitution table is synthesized anew based on the secret initial conditions of the chaotic generators, an attacker cannot exploit fixed algebraic properties. Decryption reverses this process using a corresponding inverse substitution table that maps substituted values back to their original permuted states, provided the secret parameters match perfectly.
The final processing layer applies chained XOR-based diffusion to spread the influence of each individual pixel value across the entire ciphertext matrix. In this configuration, each substituted byte is combined with its corresponding hybrid chaotic key element and the ciphertext value of the immediately preceding pixel. This recurrent dependency ensures that altering a single bit in the plain image propagates changes unpredictably across the entirety of the resulting encrypted file, producing a pronounced avalanche effect. Computational complexity analysis confirms that because each operation processes an M-by-N image matrix through sequential linear scans, the complete encryption and decryption workflows maintain an overall temporal complexity of O(MN) and a spatial storage footprint bounded strictly by the dimensions of the plain image, key matrix, cipher image, and lookup table.
Experimental validation conducted on standardized 256-by-256 pixel benchmark images, including Airplane and Peppers test files, confirmed the system’s execution speed and throughput on standard computing hardware. The proposed algorithm encrypted the Airplane benchmark image in approximately 0.007485 seconds and decrypted it in 0.012959 seconds, corresponding to throughput rates of 72.518 megabytes per second for encryption and 57.389 megabytes per second for decryption. For the Peppers test image, encryption required 0.012468 seconds at 60.763 megabytes per second, while decryption finished in 0.016631 seconds at 37.407 megabytes per second. These execution measurements verify that the non-uniform cellular automata operations and chaotic XOR combinations operate rapidly enough to handle live image feeds without causing computational bottlenecks on edge hardware.
Cryptographic quality was verified through rigorous statistical metric evaluations, including information entropy, pixel correlation, and differential sensitivity tests. Shannon entropy measurements on cipher images rose from plain values of 6.6639 for Airplane and 7.8747 for Peppers to 7.9998 and 7.9988, respectively, approaching the theoretical optimum of 8 for completely random eight-bit sequences. Adjacent pixel correlation coefficients, which ranged between 0.93 and 0.98 across red, green, and blue channels in original test images, dropped to near-zero values across horizontal, vertical, and diagonal directions after encryption. Furthermore, differential analysis yielded Number of Pixel Change Rate values of approximately 99.6 percent and Unified Average Changing Intensity scores near 33 percent, demonstrating that minor adjustments to input plaintext completely transform the resulting cipher output.
Security assessments also examined key space magnitude, key sensitivity thresholds, and resilience against common transmission impairments. Accounting for eight chaotic floating-point parameters evaluated at an operational precision of 10 to the negative 14th power, alongside the eight-byte cellular automata rule vector and the factorial combinations of the dynamic substitution box, the overall theoretical key space reaches approximately 2 to the power of 2120. This vastly exceeds the standard cryptographic threshold of 2 to the 100th power required to defeat brute-force searches. Introducing a tiny deviation of 10 to the negative 14th power into secret key values completely prevented successful decryption, yielding noise-like reconstructions. Simulated transmission tests further verified that encrypted images retained structural interpretability upon decryption when subjected to Salt-and-Pepper noise, additive Gaussian perturbations, or data cropping attacks.
Statistical randomness tests under the NIST SP 800-22 and DIEHARD evaluation suites corroborated the framework’s mathematical unpredictability, with all reported p-values exceeding the standard significance threshold of 0.01 across frequency, block frequency, runs, serial, and approximate entropy benchmarks. While the current implementation successfully demonstrates lightweight, high-speed visual data protection for static two-dimensional frames on edge devices, the authors noted that future investigations will focus on integrating the hybrid cellular automata and chaotic mapping mechanism directly into continuous real-time video streaming pipelines and adaptive artificial intelligence security architectures.
The stability of the chaotic key generator relies on bounded parameters that enforce chaotic divergence. In the one-dimensional Logistic map, selecting the control parameter within the range of 3.99 to 4 drives the system past its periodic oscillation threshold of 3.57 into full chaos, leveraging the butterfly effect where infinitesimal starting variations alter subsequent iterations. For the two-dimensional Tinkerbell map, specific numerical constants govern the coupled state equations, with parameters assigned to 0.9 for a, -0.6013 for b, 2.0 for c, and 0.50 for d to generate irregular trajectories. Within the cellular automata pipeline, each rule in the non-uniform vector corresponds to an eight-bit binary configuration covering eight possible three-cell neighbourhood states from 111 down to 000. Under this structure, Rule 51 maps directly to the binary sequence 00110011, defining exact local transition dynamics across lattice positions.
Visual and numerical assessments further confirmed how the encryption mechanism eliminates statistical vulnerabilities. Histogram evaluations revealed that while original images displayed sharp peaks reflecting pixel redundancy, the resulting cipher matrices maintained uniform distributions across all RGB channels and gray levels, preventing histogram-based statistical extraction. To evaluate distortion and recovery fidelity, the researchers evaluated Mean Squared Error alongside Peak Signal-to-Noise Ratio scaled against a maximum pixel value of 255. The encrypted images yielded high error values and low decibel ratios, confirming profound visual scrambling. Conversely, during attack simulations under realistic transmission distortions, the framework sustained an observed signal-to-noise ratio of 7.9913 decibels under noise, preserving essential structural outlines despite impulsive pixel corruptions.
Cropping attack simulations similarly demonstrated resilience against packet loss by removing portions of the cipher stream, yet decrypted outputs retained structural visual information. In comparisons with competing lightweight encryption schemes, the architecture’s key space of 2 to the power of 2120 substantially exceeded previous implementations that reached 2 to the power of 186, 2 to the power of 625, and 2 to the power of 1149. Furthermore, granular evaluations across the DIEHARD statistical suite corroborated this unpredictability, yielding p-values of 0.500 on the bitstream test and 0.083 on the serial test, verifying robust randomness across diverse AIoT operational domains.
Subject of Research: Lightweight hybrid image encryption framework for resource-constrained Artificial Intelligence of Things edge devices
Article Title: A hybrid CA-chaotic encryption framework for privacy preservation in AIoT edge devices
Article References: Yogi, B., Ghosh, N., Sahid, S., Sharma, L., & Roy, S. (2026). A hybrid CA-chaotic encryption framework for privacy preservation in AIoT edge devices. Discover Internet of Things, 6(1), Article 144. https://doi.org/10.1007/s43926-026-00472-x
Image Credits: AI Generated
DOI: 10.1007/s43926-026-00472-x
Keywords: Artificial Intelligence of Things, Cellular Automata, Chaotic Maps, Image Encryption, Edge Devices, Logistic Map, Tinkerbell Map, Dynamic S-Box, Information Entropy, Differential Cryptanalysis, NPCR, UACI, Lightweight Cryptography, Pixel Diffusion, Key Space Analysis, NIST SP 800-22
News Source: Denise Maddox. (November 13, 2026). Hybrid Cellular Automata and Chaotic Encryption Safeguards Visual Data on Edge Devices. Scienmag.



