Robust Image Watermarking in the Wavelet Domain for Copyright Protection

In this paper a new approach to image watermarking in wavelet domain is presented. The idea is to hide the watermark data in blocks of the block segmented image. Two schemes are presented based on this idea by embedding the watermark data in the low pass wavelet coefficients of each block. Due to low computational complexity of the proposed approach, this algorithm can be implemented in real time. Experimental results demonstrate the impercepti-bility of the proposed method and its high robustness against various attacks such as filtering, JPEG compres-sion, cropping, noise addition and geometric distortions.

💡 Research Summary

The paper introduces two novel image watermarking schemes that operate in the wavelet domain and are designed for real‑time copyright protection. The core idea is to segment the host image into non‑overlapping blocks, apply a 2‑D discrete wavelet transform (DWT) to each block, and embed binary watermark bits into the low‑pass coefficients of the finest scale. The embedding modifies the selected coefficients according to simple linear equations that involve a strength factor α; a value of α determines how much the coefficient is increased (for a ‘1’) or decreased (for a ‘0’). After modification, the inverse DWT reconstructs the watermarked image.

Method 1 applies this process to every block, using Daubechies‑8 symlet filters and a decomposition depth of log₂(N) for an N×N block. The strength factor is set to α = 0.1, which yields high visual fidelity (average PSNR ≈ 46 dB).

Method 2 adds a block‑selection stage: the variance of each block is computed, and only the N blocks with the highest variance are used for embedding. Because high‑variance blocks can tolerate larger modifications without perceptual degradation, α is increased to 0.25, resulting in even higher robustness while maintaining comparable PSNR (≈ 47 dB).

Detection is non‑blind: both the original and possibly attacked watermarked images are segmented and transformed in the same way. A ratio of the received to original low‑pass coefficients is computed for each block; if the majority of ratios exceed a threshold derived from α, the bit is decoded as ‘1’, otherwise ‘0’. This threshold, originally derived for an audio watermarking scheme, is reused here.

The authors evaluate the schemes on four standard 512×512 test images (Goldhill, Baboon, Barbara, Boat). They report PSNR values confirming imperceptibility and use Bit‑Error‑Rate (BER) to quantify robustness against a wide range of attacks: JPEG compression (quality 10 %–90 %), additive white Gaussian noise (σ up to 0.1), rotation (±30°), scaling (down to 0.5), and both mean and median filtering (window sizes 3×3 to 7×7). Method 2 consistently achieves lower BER than Method 1, especially under JPEG compression and filtering, while both methods remain resilient to noise and moderate geometric distortions.

A comparative study against seven established watermarking techniques—including wavelet‑based, spread‑spectrum, holographic, EPCM, multistage vector quantization, P&Z, and K&R—shows that the proposed methods outperform the benchmarks in both correlation coefficient (for JPEG) and BER (for other attacks).

Computational complexity is modest: on a 3 GHz Pentium IV, Method 1 requires 6.17 seconds and Method 2 2.66 seconds for embedding a 512×512 image; detection takes 3.92 seconds and 1.49 seconds respectively. These timings demonstrate suitability for real‑time applications.

The paper acknowledges two main limitations: reliance on a non‑blind detector (original image must be available) and a simplistic block‑selection criterion based solely on variance. Future work is suggested to develop a blind detection scheme, incorporate more sophisticated perceptual models for block selection, and possibly leverage machine‑learning techniques to further improve robustness and capacity.

In summary, the authors present a low‑complexity, high‑robustness, wavelet‑domain watermarking framework that achieves excellent visual quality and strong resistance to common image processing attacks, making it a practical candidate for real‑time copyright protection systems.