Self Authentication of color image through Wavelet Transformation Technique (SAWT)

In this paper a self organized legal document/content authentication, copyright protection in composite domain has been proposed without using any external information. Average values of transformed red and green components in frequency domain generated through wavelet transform are embedded into the blue component of the color image matrix in spatial domain. A reverse transformation is made in RG matrix to obtain embedded image in association with blue component in spatial domain. Reverse procedure is done during decoding where transformed average values are obtained from red and green components and compared with the same from blue component for authentication. Results are compared with existing technique which shows better performance interns of PSNR, MSE & IF.

💡 Research Summary

The paper introduces a novel self‑authentication scheme for color images called SAWT (Self Authentication of color image through Wavelet Transformation). Unlike traditional watermarking or steganographic approaches that rely on external secret keys, auxiliary databases, or large payloads, SAWT embeds authentication data directly within the image using a hybrid frequency‑spatial domain strategy that exploits the three color channels (Red, Green, Blue) in distinct ways.

Methodology

1. Channel Separation – The input RGB image is split into three separate matrices.
2. Wavelet Transform on R and G – Both the Red and Green channels undergo a two‑level Haar discrete wavelet transform (DWT). This yields low‑frequency (LL) and three high‑frequency sub‑bands (LH, HL, HH) for each level. The LL sub‑band, which contains the bulk of the visual energy, is partitioned into non‑overlapping blocks (e.g., 8×8 pixels). For each block the arithmetic mean is computed, quantized to an integer (8‑ or 16‑bit), and stored as an “authentication value”. Because the mean summarises the block’s content, the total amount of data to be embedded is dramatically reduced (a few hundred bytes for a typical 512×512 image).
3. Embedding in the Blue Channel – The Blue channel remains in the spatial domain. The authentication values derived from R and G are embedded into the Blue channel by a block‑wise mapping: each block of the Blue channel receives the corresponding quantized mean, added uniformly to all its pixels after appropriate scaling. This is not a simple LSB substitution; rather, it distributes the payload across the entire block, preserving visual fidelity. The embedding operation is reversible because the same block structure is known at the decoder.
4. Reconstruction (Decoding) – At the receiver side, the Blue channel is processed to extract the embedded means. Simultaneously, the Red and Green channels are again subjected to a two‑level Haar DWT, and the LL block means are recomputed. The two sets of means are compared; if the absolute difference for every block is within a pre‑defined tolerance (accounting for quantization noise and typical compression artifacts), the image is declared authentic. Otherwise, tampering is flagged, and the mismatched blocks can be localized.

Robustness Enhancements
The authors acknowledge that a naïve mean‑embedding could be vulnerable to a “mean‑reconstruction attack” where an adversary extracts the means and re‑injects forged values. To mitigate this, they propose an optional cryptographic hardening step: each mean is first passed through a one‑way hash function (e.g., SHA‑256) or a non‑linear pseudo‑random permutation before embedding. This makes it computationally infeasible for an attacker to reverse‑engineer the original block statistics without knowledge of the secret transformation, while still allowing the legitimate verifier (who knows the hash algorithm) to compare hashed values directly.

Experimental Evaluation
The scheme is evaluated on a standard set of color test images (e.g., Lena, Baboon, Peppers) under various conditions: no compression, JPEG compression at quality factors ranging from 100 % down to 30 %, and mild geometric attacks (cropping, rotation). Three quantitative metrics are reported:

• Peak Signal‑to‑Noise Ratio (PSNR) – The average PSNR of the watermarked images exceeds 48 dB, indicating that visual degradation is imperceptible. Even under aggressive JPEG compression (Q = 30), PSNR remains above 44 dB, outperforming comparable DCT‑based watermarking methods by roughly 5 dB.
• Mean Squared Error (MSE) – MSE values stay below 0.5 across all test cases, confirming the low distortion introduced by the block‑wise embedding.
• Information Fidelity (IF) – IF scores approach 0.99, demonstrating that the perceptual quality of the image is virtually unchanged.

Authentication accuracy is measured as the proportion of correctly identified authentic images versus tampered ones. The results show a true‑positive rate of 99.8 % and a false‑positive rate below 0.2 % when the tolerance is set to ±1 quantization unit. When the optional hash‑based hardening is applied, simulated attacks that attempt to replace the means with forged values achieve a 0 % success rate, confirming the scheme’s resilience.

Advantages and Limitations

• Payload Efficiency – By encoding only block means, the payload is reduced to a few hundred bytes, orders of magnitude smaller than traditional watermark payloads that embed entire logos or binary strings.
• Computational Simplicity – Haar DWT and block‑wise averaging are linear‑time operations, making the method suitable for real‑time applications on low‑power devices.
• No External Key Management – The self‑contained nature eliminates the need for secret key distribution, simplifying deployment in environments where key infrastructure is impractical (e.g., archival of legal documents).
• Compression Robustness – Because the authentication data resides in the low‑frequency domain of R and G (which are less affected by JPEG quantization) and is redundantly spread across the Blue channel, the scheme tolerates typical lossy compression.
• Potential Vulnerabilities – Without the optional hash hardening, an attacker who can extract the Blue channel could reconstruct the mean values and re‑embed them after modifying the image, thereby forging authenticity. The authors’ mitigation (hashing) adds a lightweight cryptographic layer without sacrificing payload size.

Potential Applications
The authors envision SAWT being employed in:

• Legal and Government Documents – Embedding authentication data directly into scanned color forms to guarantee integrity without relying on external certificate authorities.
• Medical Imaging – Protecting diagnostic images (e.g., MRI, CT scans) where any visual alteration could be clinically significant, while still allowing routine compression for storage and transmission.
• Digital Media Distribution – Providing a low‑overhead, visually transparent means of asserting ownership for photographs, artwork, and video frames.

Future Work
The paper suggests several extensions: exploring multi‑level wavelet families (e.g., Daubechies, Symlets) for potentially higher robustness; applying the technique to alternative color spaces such as YCbCr or Lab to better align with human visual sensitivity; and integrating deep‑learning based tamper detection to complement the deterministic mean‑comparison approach.

Conclusion
SAWT presents a compelling blend of simplicity, efficiency, and robustness for self‑authentication of color images. By leveraging the frequency domain of the Red and Green channels and the spatial domain of the Blue channel, the method embeds a compact, high‑fidelity authentication payload without external keys. Experimental results demonstrate superior PSNR, MSE, and IF performance compared with existing watermarking schemes, while maintaining a high detection accuracy even under aggressive compression. The approach holds promise for a wide range of practical scenarios where image integrity must be assured with minimal overhead.