Robust Watermarking in Multiresolution Walsh-Hadamard Transform

Gaurav Bhatnagar, Balasubramanian RamanarXiv 2009 · DOI · OpenReview

TL;DR

This paper introduces a novel multiresolution Walsh-Hadamard Transform (MR-WHT) for image processing and proposes a robust watermarking scheme for copyright protection. By embedding watermark information within the middle singular values of high-frequency sub-bands of the MR-WHT transformed image, the scheme demonstrates strong resilience against various attacks while maintaining good visual quality.

What Problem Does This Paper Address?

In the digital age, the ease of duplication and distribution of multimedia content poses significant challenges for copyright protection. Digital watermarking offers a solution by embedding hidden information within an image. While transform-domain watermarking techniques like DFT, DCT, and DWT are popular due to their robustness, this paper focuses on enhancing the Walsh-Hadamard Transform (WHT) by introducing a multiresolution version (MR-WHT). The authors aim to leverage the computational simplicity of WHT and the localization properties of multiresolution analysis, combined with Singular Value Decomposition (SVD), to create a watermarking scheme that is both visually imperceptible and highly resistant to common image manipulations and attacks.

Methods

The core of the proposed method lies in two main components:

  • Multiresolution Walsh-Hadamard Transform (MR-WHT): This is an extension of the standard WHT. It involves performing a WHT on the image, followed by specific row and column modifications. These modifications are applied iteratively to create different resolution levels, similar to how wavelet transforms decompose an image into sub-bands (LL, LH, HL, HH). The paper details a three-step process for both forward and inverse MR-WHT.
  • Singular Value Decomposition (SVD) for Watermarking: SVD is a powerful matrix factorization technique that decomposes an image into its singular values and singular vectors. Singular values are known to be robust to image manipulations. The proposed scheme applies SVD to the high-frequency (HH) sub-bands obtained from the coarsest and finest levels of the MR-WHT decomposition. The watermark is then embedded by modifying the middle singular values of these HH sub-bands using the singular values of the watermark image. The authors argue that middle singular values offer a good balance between image quality (largest singular values are crucial) and noise sensitivity (smallest singular values are most affected).

The embedding process involves:

  1. Applying L-level MR-WHT to the host image.
  2. Selecting the HH sub-band from the coarsest (level 1) and finest (level L) levels.
  3. Performing SVD on these HH sub-bands and the watermark.
  4. Modifying the middle singular values of the HH sub-bands with those of the watermark, scaled by a watermark strength factor ($\alpha$) and a factor ($\rho$) that determines which singular values are modified.
  5. Reconstructing the modified HH sub-bands using inverse SVD.
  6. Mapping these modified sub-bands back to their original positions and applying inverse MR-WHT to obtain the watermarked image.

The extraction process is similar but in reverse, using the host and watermarked images to recover the watermark.

Datasets

The paper evaluates the proposed scheme using standard grayscale images:

  • Host Images: Payaso, Yacht, and Zelda (512x512 pixels).
  • Watermark Images: Peacock, Cup, and IEEE CS (64x64 pixels).

The authors do not introduce new benchmarks but rather use existing standard images for evaluation.

Key Insights and Ideas

  • MR-WHT for Enhanced Localization: By extending WHT to a multiresolution framework, the scheme gains the ability to decompose the image into different frequency sub-bands, allowing for more targeted embedding and extraction.
  • Middle Singular Values for Robustness and Quality: The choice to embed watermarks in the middle singular values of the HH sub-bands is a key insight. This strategy aims to strike a balance, preserving image quality by not altering the most significant singular values and maintaining robustness by avoiding the most noise-sensitive ones.
  • SVD's Inherent Robustness: The paper leverages the well-established robustness of SVD in watermarking, where singular values are less affected by common image processing operations compared to pixel values.
  • Computational Simplicity: The underlying WHT is computationally efficient, making the MR-WHT and subsequent watermarking operations potentially faster than methods relying on more complex transforms.

Key Results

The experimental results demonstrate the effectiveness of the proposed watermarking scheme:

  • Visual Imperceptibility: The paper claims "better visual imperceptibility," and the figures show that the watermarked images appear very similar to the originals.
  • Robustness: The scheme shows significant resilience against various attacks, including:
  • Noise: 100% Gaussian Noise and 100% Salt & Paper Noise.
  • Compression: JPEG Compression (CR=100).
  • Geometric Attacks: Row & Column Deletion, Pixelation, Cropping (2.5% area remaining), Vertical Flipping, Horizontal Flipping, Sharpening, and Wrapping.
  • Quantitative Metrics: Correlation coefficients between the original and extracted watermarks are reported. For most attacks, these coefficients are high (close to 1 or -1), indicating successful watermark recovery. For example, after JPEG compression, correlation coefficients are around 0.48-0.56. For cropping, they are very high, around -0.99.
  • PSNR: The Peak Signal-to-Noise Ratio (PSNR) values for the watermarked images are generally high (e.g., 48.7463 dB for Payaso), suggesting good image quality.

Limitations and Future Work

  • Limitations:
    • The paper mentions that for rectangular images, they are made square by adding zero-filled rows/columns, which might introduce some artifacts or affect efficiency.
  • The robustness against certain attacks, like Gaussian Blur (13x13), shows negative correlation coefficients (e.g., -0.2704), indicating that while the watermark is detected, its orientation might be inverted or the correlation is weak.
  • The paper focuses on grayscale images.
  • Future Work: The authors suggest that the proposed algorithm stands with various attacks, implying that further exploration of its robustness against a wider range of attacks and optimizations could be beneficial. The use of $\rho$ as a key for extraction adds complexity for attackers, suggesting that exploring different key management strategies could be a future direction.

Summary updated

Abstract

In this paper, a newer version of Walsh-Hadamard Transform namely multiresolution Walsh-Hadamard Transform (MR-WHT) is proposed for images. Further, a robust watermarking scheme is proposed for copyright protection using MRWHT and singular value decomposition. The core idea of the proposed scheme is to decompose an image using MR-WHT and then middle singular values of high frequency sub-band at the coarsest and the finest level are modified with the singular values of the watermark. Finally, a reliable watermark extraction scheme is developed for the extraction of the watermark from the distorted image. The experimental results show better visual imperceptibility and resiliency of the proposed scheme against intentional or un-intentional variety of attacks.
BibTeX
@misc{bhatnagar2009robust,
  author = {Gaurav Bhatnagar and Balasubramanian Raman},
  title = {{Robust Watermarking in Multiresolution Walsh-Hadamard Transform}},
  year = {2009},
  eprint = {0906.4415},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/0906.4415},
}
Figure