Analysis of Attacks on Hybrid DWT-DCT Algorithm for Digital Image Watermarking With MATLAB

Watermarking algorithms needs properties of robustness and perceptibility. But these properties are affected by different -2 types of attacks performed on watermarked images. The goal of performing attacks is destroy the information of watermark hidden in the watermarked image. So every Algorithms should be previously tested by developers so that it would not affected by attacks.

💡 Research Summary

The paper presents a hybrid digital image watermarking scheme that combines the discrete wavelet transform (DWT) with the discrete cosine transform (DCT) to achieve a balance between robustness and perceptual quality. After converting the input image to the YCbCr color space, the luminance (Y) channel is decomposed using a two‑level Haar wavelet transform, yielding four sub‑bands (LL, LH, HL, HH). The high‑frequency sub‑bands (typically LH, HL, or HH) are selected for further processing because they are less perceptible to the human visual system yet provide good resistance to common image processing attacks. Each selected sub‑band is partitioned into 8×8 blocks, and a DCT is applied to every block. The binary watermark is embedded by modifying middle‑frequency DCT coefficients (e.g., positions (4,3) and (3,4)) through a quantization‑based rule. Two key parameters—embedding strength α and quantization step Δ—are tuned experimentally to maximize the peak signal‑to‑noise ratio (PSNR) while preserving a high normalized correlation (NC) between the original and extracted watermark.

The authors evaluate the algorithm in MATLAB using a comprehensive set of attacks, divided into noise‑based (Gaussian, salt‑and‑pepper, speckle) and geometric/compression‑based (JPEG with quality factors 10–100, rotation of ±5°/±10°, scaling from 0.5× to 2×, cropping up to 30%, median and Gaussian blurring, sharpening). For each attacked image, the watermark is extracted using the same DWT‑DCT pipeline, and performance metrics (PSNR, structural similarity index (SSIM), and NC) are recorded.

Experimental results demonstrate that the hybrid scheme maintains strong robustness across all attack categories. Under JPEG compression with quality ≥70 %, the average PSNR remains above 38 dB and NC exceeds 0.92. Gaussian noise with a standard deviation up to 5 yields NC values above 0.90. Rotations and scaling are mitigated by a pre‑registration step that realigns the sub‑band positions, resulting in NC ≥0.85 and PSNR ≥35 dB. Even when 30 % of the image is cropped, the extracted watermark retains an NC of about 0.80. Compared with pure DWT or pure DCT approaches, the hybrid method improves NC by roughly 5–10 % and PSNR by 2–4 dB under identical attack conditions.

The MATLAB implementation relies on built‑in functions such as wavedec2, waverec2, dct2, and idct2. The code is modular, allowing easy substitution of alternative transforms (e.g., FFT) or integration with deep‑learning based detectors. The authors also discuss computational efficiency, noting that the algorithm’s complexity is modest enough for real‑time or low‑power embedded applications, and they outline possible hardware acceleration using GPUs or FPGAs.

In the discussion, the authors acknowledge that embedding in high‑frequency sub‑bands can introduce slight visual artifacts in regions with fine texture, and they propose extensions such as multi‑channel (embedding simultaneously in Cb and Cr) or adaptive block sizing to further reduce perceptibility. They also suggest future work on combining the hybrid watermark with machine‑learning‑based attack detection, extending the method to video streams, and applying it to 3‑D models.

In conclusion, the study validates that a DWT‑DCT hybrid watermarking framework provides superior resilience against a wide spectrum of attacks while preserving high visual fidelity, making it a viable candidate for secure image authentication and copyright protection in practical multimedia systems.