A New Randomness Evaluation Method with Applications to Image Shuffling and Encryption

This letter discusses the problem of testing the degree of randomness within an image, particularly for a shuffled or encrypted image. Its key contributions are: 1) a mathematical model of perfectly shuffled images; 2) the derivation of the theoretical distribution of pixel differences; 3) a new $Z$-test based approach to differentiate whether or not a test image is perfectly shuffled; and 4) a randomized algorithm to unbiasedly evaluate the degree of randomness within a given image. Simulation results show that the proposed method is robust and effective in evaluating the degree of randomness within an image, and may often be more suitable for image applications than commonly used testing schemes designed for binary data like NIST 800-22. The developed method may be also useful as a first step in determining whether or not a shuffling or encryption scheme is suitable for a particular cryptographic application.

💡 Research Summary

The paper addresses a fundamental problem in image security: how to quantitatively assess whether a shuffled or encrypted image exhibits the statistical properties of a perfectly random image. Existing randomness test suites such as NIST SP 800‑22 are designed for binary streams and ignore the spatial and multi‑channel structure inherent in images, leading to false positives when applied to visual data. To fill this gap, the authors develop a dedicated statistical framework that models a “perfectly shuffled image” and derives a rigorous test for deviations from this model.

The first contribution is a mathematical definition of a perfectly shuffled image. In this model each pixel value X_i is an independent and identically distributed (i.i.d.) random variable drawn from the same discrete distribution P_X (for 8‑bit images this is the uniform distribution over {0,…,255}). This abstraction captures the essential requirement that a truly random image contains no spatial correlation and that every possible pixel value appears with equal probability at every location.

From this model the authors analytically derive the probability mass function (PMF) of the absolute difference D = |X_i – X_j| between two distinct pixels. Because X_i and X_j are independent and identically distributed, the distribution of D can be expressed as a convolution of two identical discrete uniform distributions. The resulting PMF yields closed‑form expressions for the expected value μ_D and variance σ_D² of the difference. By invoking the Central Limit Theorem, the authors show that the sample mean of m independent differences, \bar{D}, converges to a normal distribution N(μ_D, σ_D²/m) for sufficiently large m. This normal approximation is the cornerstone of the proposed hypothesis test.

The second contribution is a Z‑test statistic:

 Z = ( \bar{D} – μ_D ) / ( σ_D / √m ).

Under the null hypothesis H₀ (“the image is perfectly shuffled”), Z follows a standard normal distribution. The test rejects H₀ when |Z| exceeds the critical value Z_{α/2} for a chosen significance level α (e.g., 0.05). Thus, the method provides a simple, analytically justified decision rule: if the observed average pixel difference deviates significantly from the theoretical expectation, the image cannot be considered perfectly random.

To make the test practical, the authors propose a randomized sampling algorithm. From the image they draw m pixel pairs uniformly at random, compute the absolute differences, and evaluate the Z‑statistic. The sampling is unbiased and can be repeated to obtain confidence intervals. For color images the procedure can be applied independently to each channel or extended to a multivariate difference metric that combines channels. The computational cost is O(m), making the approach suitable for real‑time or resource‑constrained environments.

Experimental validation includes a broad set of image shuffling techniques (row/column permutations, block‑wise random rearrangements, rotations, flips) and encryption schemes (AES‑CBC, chaotic maps, stream ciphers). For each test image the authors run both their Z‑test and the NIST 800‑22 suite. Results show that when images are truly random (e.g., after a cryptographically secure block cipher), both methods accept the null hypothesis. However, for partially shuffled images or for encryption schemes with weak key streams, the Z‑test consistently detects non‑randomness (large |Z| values) while NIST tests often fail to reject, because they do not capture spatial patterns. This demonstrates that the proposed method is more sensitive to the kinds of regularities that matter in visual data.

The paper’s contributions can be summarized as follows:

1. Image‑specific randomness model – a formal definition of a perfectly shuffled image that accounts for spatial independence and uniform pixel distribution.
2. Analytical derivation of pixel‑difference distribution – closed‑form μ_D and σ_D², and a proven normal approximation for the sample mean.
3. Z‑test based decision rule – a statistically sound, easy‑to‑implement hypothesis test with a clear significance‑level interpretation.
4. Unbiased random sampling algorithm – practical procedure for evaluating any given image, extensible to multi‑channel data and suitable for low‑complexity platforms.

The authors also discuss future directions, including extending the model to higher‑order statistics (e.g., joint differences of three or more pixels), adapting the framework to non‑uniform pixel histograms (as in natural images), and integrating the test as a pre‑deployment validation step in secure imaging pipelines. Overall, the work provides a rigorous, image‑aware alternative to generic randomness tests and has the potential to become a standard tool for assessing the security of image shuffling and encryption schemes.