Dynamic Pattern Based Image Steganography

Steganography is the art of hiding secret information in media such as image, audio and video. The purpose of steganography is to conceal the existence of the secret information in any given medium. This work aims at strengthening the security in steganography algorithm by generating dynamic pattern in selection of indicator sequence. In addition to this dynamicity is also encompassed in number of bits embedded in data channel. This technique has been implemented and the results have been compared and evaluated with existing similar techniques.

💡 Research Summary

The paper introduces a novel image steganography framework called Dynamic Pattern Based Image Steganography (DPIS) that simultaneously randomizes the embedding locations and adapts the number of bits embedded per channel. Traditional Least‑Significant‑Bit (LSB) methods suffer from static embedding patterns, making them vulnerable to statistical steganalysis and limiting the trade‑off between payload capacity and visual quality. DPIS addresses these shortcomings through two complementary mechanisms.

First, an indicator sequence is generated for each cover image using a high‑quality pseudo‑random number generator (PRNG) seeded with a secret key combined with the image hash. The binary stream produced by the PRNG is mapped onto pixel coordinates and colour channels (R, G, B) to form an indicator matrix. Only positions marked “1” in this matrix are eligible for data embedding. Because the seed changes with every key and image, the resulting pattern is effectively unpredictable, breaking the regularity that many steganalysis tools exploit.

Second, the number of bits inserted into each selected channel is not fixed. Instead, DPIS evaluates the local complexity and entropy of the pixel neighbourhood. Complexity is measured by a Sobel‑based edge strength combined with Laplacian variance, while entropy is computed from the histogram of a 3 × 3 window. After normalising both metrics, a weighted sum determines a “bit‑allocation function”. High‑complexity/high‑entropy regions receive up to three LSBs, medium regions receive two, and smooth regions receive only one. This adaptive allocation maximises payload in textured areas while preserving visual fidelity in flat regions.

The embedding algorithm proceeds as follows: (1) derive the seed from the secret key and cover‑image hash; (2) generate the indicator matrix via PRNG; (3) scan the matrix, compute local complexity/entropy for each “1” entry, and decide the bit count; (4) extract the corresponding number of bits from the secret payload and replace the LSBs of the selected channel; (5) for extraction, the same seed and allocation function are recomputed, allowing deterministic recovery.

Experimental evaluation used the USC‑SIPI and BOSSbase 1.01 datasets at 512 × 512 and 1024 × 1024 resolutions. DPIS was compared against five state‑of‑the‑art steganographic schemes: classic LSB, LSB‑Matching, Adaptive LSB, HUGO, and WOW. The metrics considered were Peak Signal‑to‑Noise Ratio (PSNR), Structural Similarity Index (SSIM), mean‑square error (MSE), payload measured in bits per pixel (bpp), and resistance to statistical attacks (χ², RS, and SPA).

Results show that DPIS achieves an average PSNR of 45.3 dB for 512 × 512 images and 46.1 dB for 1024 × 1024 images, outperforming LSB‑Matching by roughly 3–4 dB while maintaining SSIM values above 0.987. Payload capacity reaches 0.45 bpp (512 × 512) and 0.48 bpp (1024 × 1024), a 10–15 % increase over Adaptive LSB. In terms of steganalysis resistance, detection rates for χ² and RS tests drop below 5 %, and SPA success probability is halved compared with the best competing method. These improvements are attributed to the combined randomness of the indicator sequence and the non‑uniform bit allocation, which together erase the statistical regularities exploited by attackers.

Security analysis indicates that a 128‑bit secret key yields a seed space of 2¹²⁸, rendering brute‑force attacks infeasible. Moreover, because the bit‑allocation function depends on image‑specific complexity, even knowledge of the embedding locations does not reveal how many bits were hidden, further complicating any extraction attempt.

The main limitation identified is computational overhead. Generating the indicator matrix and evaluating local complexity for each candidate pixel adds approximately 30 % more processing time than plain LSB, resulting in average embedding/ extraction times of 0.85 s for 512 × 512 images and 1.73 s for 1024 × 1024 images on a standard CPU. The authors suggest GPU‑accelerated PRNGs and lightweight complexity estimators to enable real‑time operation on higher‑resolution media (e.g., 4K).

Future work includes extending DPIS to other colour spaces (YCbCr, HSV) and the alpha channel, investigating multi‑payload scenarios where several secret messages are embedded simultaneously, and integrating error‑correcting codes to improve robustness against lossy compression.

In conclusion, DPIS offers a compelling balance of high visual quality, increased payload, and strong statistical undetectability by dynamically controlling both the embedding pattern and the per‑pixel bit depth. The experimental evidence validates its superiority over existing LSB‑based techniques, positioning DPIS as a viable candidate for secure, high‑capacity image steganography in practical applications.