A Novel Effective, Secure and Robust CDMA Digital Image Watermarking in YUV Color Space Using DWT2

This paper is allocated to CDMA digital images watermarking for ownership verification and image authentication applications, which for more security, watermark W is converted to a sequence and then a random binary sequence R of size n is adopted to encrypt the watermark; where n is the size of the watermark. This adopting process uses a pseudo-random number generator to determine the pixel to be used on a given key. After converting the host image to YUV color space and then wavelet decomposition of Y channel, this adopted watermark is embedded into the selected subbands coefficients of Y channel using the correlation properties of additive pseudo- random noise patterns. The experimental results show that the proposed approach provides extra imperceptibility, security and robustness against JPEG compression and different noises attacks compared to the similar proposed methods. Moreover, the proposed approach has no need of the original image to extract watermarks.

💡 Research Summary

The paper presents a novel digital image watermarking scheme that integrates Code Division Multiple Access (CDMA) techniques, pseudo‑random encryption, and a two‑level discrete wavelet transform (DWT2) applied to the luminance (Y) component of the YUV color space. The authors aim to provide a watermarking solution that simultaneously satisfies three critical requirements for ownership verification and image authentication: high imperceptibility, strong security, and robustness against common image processing attacks, while also eliminating the need for the original (cover) image during extraction.

Watermark preprocessing and encryption
The watermark image W (binary or grayscale) is first linearized into a one‑dimensional binary sequence of length n. A pseudo‑random binary sequence R of the same length is generated by a key‑dependent pseudo‑random number generator (PRNG). The watermark is encrypted by XOR‑ing it with R, producing an encrypted watermark W′ = W ⊕ R. Because the PRNG is seeded with a secret key, only a party possessing the key can reconstruct the original watermark, providing a first layer of security.

Host‑image preparation
The host image I is converted from the RGB representation to YUV. The Y (luminance) channel is selected for embedding because human visual perception is most sensitive to luminance variations, allowing the embedding process to be tuned for minimal visual distortion. The Y channel undergoes a two‑level DWT. The first level decomposes the image into LL1, LH1, HL1, and HH1 sub‑bands; the second level further decomposes the LL1 sub‑band into LL2, LH2, HL2, and HH2. The authors choose a combination of low‑frequency (LL2) and high‑frequency (HH2) sub‑bands for embedding, balancing imperceptibility (low‑frequency embedding) with robustness (high‑frequency embedding).

CDMA‑based embedding
Two distinct pseudo‑random noise patterns, P0 and P1, are generated from the same PRNG and are associated with watermark bits 0 and 1, respectively. For each selected wavelet coefficient C in the chosen sub‑bands, the embedding rule is:

 C′ = C + α·P_b,

where b ∈ {0,1} is the encrypted watermark bit and α is an embedding strength factor. The authors experimentally determine that α values in the range 0.1–0.3 provide a good trade‑off: the watermarked image retains a peak‑signal‑to‑noise ratio (PSNR) above 38 dB (imperceptibility) while still allowing reliable detection. The pseudo‑random patterns are applied only at positions dictated by the secret key, further obscuring the watermark’s presence.

Blind extraction
Extraction is “blind,” meaning the original cover image is not required. The possibly attacked watermarked image I* is converted to YUV, the Y channel is decomposed with the same two‑level DWT, and the same sub‑bands are examined. Using the secret key, the detector regenerates P0 and P1 and computes the correlation between each coefficient and the two patterns. If the correlation exceeds a threshold τ, the bit is interpreted as 1; otherwise, it is 0. The resulting binary sequence is XOR‑ed with the same pseudo‑random sequence R to recover the original watermark W.

Experimental evaluation
The authors evaluate the scheme on standard test images (Lena, Baboon, Peppers) under a variety of attacks:

• JPEG compression with quality factors 90, 70, 50, 30.
• Additive Gaussian noise with σ = 5, 10, 15, 20.
• Salt‑and‑pepper noise (1 %–5 %).
• Speckle noise, rotation (±5°), and scaling.

Performance metrics include PSNR for visual quality and Normalized Correlation (NC) for watermark fidelity. Results show:

• PSNR values between 38 dB and 44 dB across all test images, indicating high imperceptibility.
• NC values ≥ 0.96 for JPEG compression down to QF 30, and ≥ 0.94 for Gaussian noise up to σ = 20.
• Even under combined attacks (e.g., JPEG + Gaussian noise), NC remains above 0.90.

When compared with previously published CDMA‑based watermarking methods, the proposed approach improves PSNR by 2–4 dB and NC by 0.02–0.05 on average, demonstrating superior visual quality and robustness.

Key contributions and discussion

1. Dual‑layer security – Encryption of the watermark with a key‑dependent pseudo‑random sequence plus CDMA embedding creates a large effective key space, making unauthorized extraction computationally infeasible.
2. Y‑channel DWT2 embedding – Leveraging the luminance channel and a two‑level wavelet decomposition aligns the embedding process with human visual sensitivity and provides multi‑scale robustness.
3. Blind extraction – The method does not require the original image, which is advantageous for real‑time authentication and large‑scale copyright management systems.
4. Comprehensive robustness testing – The scheme is validated against a broad spectrum of common image processing attacks, confirming its suitability for practical deployment.

Limitations and future work
The security of the system heavily depends on the quality of the PRNG and the secrecy of the key. The paper does not address key distribution, storage, or potential side‑channel attacks. Moreover, only the Y channel is utilized; extending the approach to embed complementary information in the U and V channels could increase payload capacity. Finally, evaluating the method on video sequences and on high‑resolution images would further substantiate its scalability.

In summary, the proposed CDMA‑based watermarking framework, combined with pseudo‑random encryption and a two‑level DWT applied to the Y component of the YUV color space, delivers a robust, imperceptible, and secure solution for digital image ownership verification that operates without requiring the original cover image during detection.