💻 computer science

Image Encryption via Data-Identified Discrete Chaotic Maps

This paper proposes a novel data-driven image encryption framework that utilizes the SINDy-PI algorithm to learn explicit chaotic map dynamics directly from observational data, thereby creating a highly secure, data-dependent encryption scheme that outperforms traditional fixed-map methods in sensitivity, randomness, and resistance to statistical and differential attacks.

Original authors: Wenyuan Lia, Xiao-Yun Wang, Zhigang Zhu, Xiaofeng Zhang, Li Zhang

Published 2026-05-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Wenyuan Lia, Xiao-Yun Wang, Zhigang Zhu, Xiaofeng Zhang, Li Zhang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: A Lock That Builds Itself

Imagine you want to send a secret photo to a friend. Usually, you use a lock (an encryption algorithm) that everyone knows how to build, and you just give them a specific key (like a password) to open it. If a hacker steals the key, they can open the lock.

This paper proposes a completely different way. Instead of giving your friend a key to a pre-made lock, you give them a recipe and a bag of ingredients. They use those ingredients to build a unique lock right in front of them.

The "ingredients" are the data used to train the system, and the "recipe" is a smart computer algorithm called SINDy-PI. The "key" is just the starting point (two tiny numbers).

How It Works: The "Shape-Shifting" Lock

1. The Smart Detective (SINDy-PI)
Think of the SINDy-PI algorithm as a super-smart detective. Usually, scientists have to guess what the rules of a chaotic system (a complex, unpredictable machine) are. This detective doesn't guess. It looks at a pile of data (observations of the machine moving) and figures out the exact mathematical rules that govern it. It can even find hidden rules that humans might miss, like complex interactions between different parts of the machine.

2. The Data-Dependent Map
In this new system, the "lock" isn't a fixed shape. It's a chaotic map (a mathematical rule that scrambles numbers).

Old Way: Everyone uses the same map (like the famous Hénon map).
New Way: The map is built from the data. If you feed the detective slightly different data (even data with a tiny bit of random "noise," like static on a radio), the detective builds a slightly different map.

3. The Secret Key
The only thing you need to share with your friend to unlock the message is the starting point (two numbers, like 0.2 and 0.3).

The Twist: Even if a hacker steals those two numbers, they can't open the lock. Why? Because they don't know which map was built. Did the map have a tiny extra term? Was it built from data with a specific type of noise? Without the exact map structure, the key is useless.

The Process: Scrambling the Image

Once the unique map is built, here is how the photo is scrambled:

Shuffling: The map generates a chaotic sequence of numbers. These numbers tell the computer to shuffle the rows and columns of the image, like mixing a deck of cards.
Diffusing: The map generates more numbers to change the actual colors of the pixels. If you change just one pixel in the original photo, the map ensures that every single pixel in the final scrambled photo changes. It's like dropping a single drop of red ink into a bucket of water; the whole bucket turns pink.

Why It's So Secure (The Results)

The authors tested this on three different types of chaotic systems (the Hénon map, a 3D Logistic map, and a Lozi map) and found it works great for all of them. Here is what their tests showed:

The "One Drop" Test (Sensitivity): If you change the starting numbers by an amount so small it's almost impossible to measure (0.0000000000000001), the decryption fails completely. The result is just random noise. This proves the system is incredibly sensitive to the key.
The "Statistical Camouflage" (Histograms): A normal photo has patterns (lots of dark pixels, few bright ones). The encrypted photo looks like pure static. If you count the pixels, they are spread out perfectly evenly, giving hackers no clues to work with.
The "Neighbor" Test (Correlation): In a normal photo, a pixel is usually very similar to the one next to it. In the encrypted photo, neighbors are completely unrelated. The connection is totally broken.
The "Tiny Change" Test (Differential Attacks): If a hacker changes just one tiny bit in the original photo, the encrypted version changes completely (about 99.6% of the pixels change). This makes it impossible to guess the original by comparing two encrypted versions.

The "Hidden" Security Layer

The most unique part of this paper is the concept of the "Implicit Key."
Usually, security relies on how many keys you have. Here, the security relies on the fact that the lock itself changes based on the training data.

Analogy: Imagine two people have the same key (the starting numbers). But Person A built their lock using "Data Set A," and Person B built their lock using "Data Set B" (which had a tiny bit of different noise). Even though the keys are identical, the locks are slightly different. If a hacker has the key but doesn't know which "Data Set" was used to build the lock, they are stuck.

Summary

This paper introduces a new way to encrypt images where the encryption rules aren't fixed. Instead, they are discovered on the fly from data using a smart algorithm. This adds a hidden layer of security: even if someone steals your password, they can't unlock your data unless they also know the exact "ingredients" used to build the lock. The system is extremely sensitive, statistically perfect, and works on various types of chaotic systems.

Technical Summary: Image Encryption via Data-Identified Discrete Chaotic Maps

Problem Statement
Traditional chaos-based image encryption schemes predominantly rely on predefined, fixed chaotic systems (e.g., the standard Hénon map or 3D logistic maps). A significant security concern with these approaches is that the mathematical structure of the chaotic map is often known or easily reverse-engineered, leaving the system vulnerable if the initial conditions (the key) are compromised. Furthermore, proposing entirely new chaotic systems manually is a complex and difficult task. The paper addresses the need for a more robust encryption paradigm where the chaotic map itself is not fixed but is dynamically derived from data, thereby adding an implicit layer of security beyond the traditional key space.

Methodology
The authors propose a data-driven image encryption framework that utilizes the SINDy-PI (Sparse Identification of Nonlinear Dynamics - Parallel Implicit) algorithm. The methodology involves the following core steps:

Algorithm Adaptation: The SINDy-PI algorithm, originally designed for continuous-time systems, is modified for discrete-time chaotic systems. This involves reconstructing the library of preselection functions to match mapping characteristics and adjusting the data generation logic to use cyclic iteration rather than time-step integration methods (like Runge-Kutta).
Data-Driven Map Identification: Instead of using a pre-defined equation, the system generates observational data from a chaotic system (e.g., Hénon, 3D logistic, or Lozi maps). The SINDy-PI algorithm then identifies the explicit dynamical equations, including cross-terms and higher-order nonlinearities, directly from this data.
Encryption Scheme:
- Key Structure: The encryption key consists solely of the initial conditions $(x_0, y_0)$ . The specific form of the chaotic map is not transmitted as a key; it is derived from the training data used to generate the map.
- Process: Using the identified map and the initial conditions, a chaotic sequence is generated. This sequence is used for:
  - Scrambling: Independent row and column sorting based on the chaotic sequence to permute pixel positions.
  - Diffusion: Multi-round (four rounds) additive diffusion (mod 256) to alter pixel values, ensuring that a single-bit change in the plaintext affects the entire ciphertext.
Decryption: Successful decryption requires the exact same initial conditions and the exact same map structure (symbolic expression) used during encryption. Since the map is data-dependent, an attacker with the correct initial conditions but the wrong map structure (derived from different training data) cannot decrypt the image.

Key Contributions

First Application to Discrete Systems: This work extends the SINDy-PI algorithm, previously limited to continuous-time systems, to the identification of discrete chaotic maps.
Data-Dependent Implicit Key: The paper introduces a novel security mechanism where the chaotic map structure itself acts as an implicit key. Even with identical initial conditions, different training data (e.g., varying noise seeds) result in slightly different map structures, producing completely different ciphertexts.
Generality: The framework is validated across three distinct chaotic systems: the 2D Hénon map, the 3D logistic map, and the piecewise-linear Lozi map, demonstrating its applicability to various dynamical structures.
Enhanced Sensitivity: The scheme exhibits extreme sensitivity to initial conditions (a perturbation of $10^{-16}$ causes total decryption failure) and high sensitivity to the map structure itself.

Experimental Results
The framework was tested using a 256 $\times$ 256 grayscale "Moon surface" image. The results indicate:

Information Entropy: The encrypted images achieved near-ideal entropy values ( $\approx 7.9976$ bits), indicating a highly uniform pixel distribution.
Correlation: Inter-pixel correlation was reduced to negligible levels (approx. 0.08 in horizontal, vertical, and diagonal directions), significantly lower than fixed-parameter schemes.
Key Sensitivity: Perturbations as small as $10^{-16}$ in the initial values resulted in a PSNR of $\approx 10.2$ dB (far below the 30 dB lossless threshold), confirming total decryption failure.
Differential Attacks: The scheme demonstrated strong resistance to differential attacks, with NPCR (Number of Pixels Change Rate) averaging 99.6556% and UACI (Unified Average Change Intensity) averaging 33.5824%, closely matching theoretical ideals.
Map Sensitivity: When the same initial conditions were used with maps identified from slightly different training data (noise-free vs. low-noise), the resulting ciphertexts showed an NPCR of 99.6231% and UACI of 33.4057%, confirming that the map structure acts as a critical, independent security factor.

Significance and Claims
The paper claims to establish a new paradigm for chaos-based cryptography by unifying system identification with encryption. The primary significance lies in moving beyond fixed maps to adaptive, data-centric security architectures. By making the map structure dependent on the training data, the authors argue that the system introduces an extra layer of security that resists reverse engineering and known-plaintext attacks more effectively than traditional schemes. The authors assert that this approach not only provides more options for algorithm design but also offers a solution to the issue of low security in existing chaos-based algorithms. Future work is outlined as extending the method to hyperchaotic systems and validating it on experimental time series from physical chaotic circuits.