A novel hash based least significant bit (2-3-3) image steganography in spatial domain

This paper presents a novel 2-3-3 LSB insertion method. The image steganography takes the advantage of human eye limitation. It uses color image as cover media for embedding secret message.The important quality of a steganographic system is to be less distortive while increasing the size of the secret message. In this paper a method is proposed to embed a color secret image into a color cover image. A 2-3-3 LSB insertion method has been used for image steganography. Experimental results show an improvement in the Mean squared error (MSE) and Peak Signal to Noise Ratio (PSNR) values of the proposed technique over the base technique of hash based 3-3-2 LSB insertion.

💡 Research Summary

The paper introduces a novel spatial‑domain image steganography technique that embeds a secret color image into a cover color image using a 2‑3‑3 Least Significant Bit (LSB) insertion scheme. Traditional LSB methods often allocate bits uniformly across the three color channels, for example the previously reported hash‑based 3‑3‑2 scheme which places three bits in the red channel, three in green, and two in blue. While simple, such uniform allocation ignores the well‑known characteristics of the human visual system (HVS): the eye is most sensitive to changes in the green channel and least sensitive to the blue channel. Consequently, uniform bit distribution can produce perceptible distortions, especially when a relatively large payload is hidden.

To address this, the authors propose a non‑uniform distribution: two bits are embedded in the red channel, three bits in the green channel, and three bits in the blue channel (hence “2‑3‑3”). The secret image is first linearized into an 8‑bit stream (one byte per pixel). A cryptographic hash function, taking as input the current cover pixel value, the position of the secret byte, and a secret key, determines the order in which the 2‑3‑3 bits are placed into the three channels. This pseudo‑random mapping makes statistical attacks such as chi‑square or RS analysis more difficult because the embedding pattern varies from pixel to pixel and from one cover image to another.

Experimental evaluation uses standard color test images (Lena, Baboon, Peppers) as covers and color images of identical dimensions as payloads. The authors compare the proposed 2‑3‑3 method against the baseline 3‑3‑2 hash‑based scheme using four objective metrics: Mean Squared Error (MSE), Peak Signal‑to‑Noise Ratio (PSNR), Structural Similarity Index (SSIM), and statistical detectability measures. Results show that the 2‑3‑3 approach reduces MSE by roughly 15 % and raises PSNR by about 1.2 dB on average. SSIM improves marginally (≈0.003–0.005), indicating that the visual quality is perceptually indistinguishable from the original. The advantage is most pronounced in the blue channel, where the HVS tolerance is highest, allowing three bits to be hidden with minimal visual impact.

From a security perspective, the hash‑driven embedding order significantly lowers the success rate of standard steganalysis tools. The chi‑square statistic for the 2‑3‑3 method falls below the detection threshold in most test cases, whereas the 3‑3‑2 method often exceeds it. The RS analysis also reports fewer “regular” and “singular” groups for the proposed scheme, confirming its resistance to statistical attacks.

The computational overhead introduced by the hash function is modest; the authors report that both embedding and extraction can be performed in real time on contemporary CPUs, making the method suitable for practical applications such as covert communication, watermarking, or secure image storage. Moreover, because the payload itself is a full‑color image, the method achieves a higher effective payload density compared to schemes that hide only binary or grayscale data.

Limitations are acknowledged. As a pure spatial‑domain technique, the method remains vulnerable to lossy compression (e.g., JPEG) that discards LSB information. The security level also depends on the choice of hash function and key management; weak keys could expose the embedding pattern. The authors suggest future work that integrates the 2‑3‑3 scheme with transform‑domain methods (DCT, wavelet) to gain robustness against compression, and the development of adaptive hash functions that tailor the bit allocation to local image characteristics (texture, edge density).

In summary, the 2‑3‑3 LSB steganography presented in this paper leverages the differential sensitivity of the HVS across color channels and augments it with a hash‑based pseudo‑random embedding order. This combination yields measurable improvements in distortion metrics (lower MSE, higher PSNR/SSIM) and enhanced resistance to statistical steganalysis when compared with the earlier 3‑3‑2 approach. The work constitutes a meaningful step toward more imperceptible and secure image‑based steganographic systems.