A Study on the Comparative Analysis of Embedded and Zero Watermarking for Unstructured Image Protection

초록

Embedded watermarking and zero-watermarking, both of which are technologies for copyright protection of digital images, are being actively studied based on their respective advantages and disadvantages. As the prevalence of copyright infringement and the use of irregular-resolution images in game content and media applications increases, there is a growing need for practical and robust image protection techniques. In this paper, we implement discrete wavelet transform (DWT)-discrete cosine transform (DCT)-singular value decomposition (SVD)-based embedded watermarking and hash-based XORbased zero-watermarking algorithms in Python for about 200 non-standard images with various resolutions and shapes (resolution range: 512 x 512 to 6800 x 4000) and quantitatively compare and analyse their performance. We evaluate the robustness and restoration rate of each watermarking method by applying various attacks such as JPEG compression, blur, Gaussian noise, cropping, and rotation based on peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and normalized correlation (NC) indices. The experimental results show that the embedded method showed fast processing speed and stable quality maintenance performance even in high-resolution images, and zero watermarking had the advantage of not damaging the original image, but was relatively prone to being affected by restoration sensitivity and execution time. While not directly implemented in this study, the findings provide a foundational reference for integrating robust watermarking mechanisms into blockchain and InterPlanetary File System (IPFS) based copyright authentication systems, particularly for protecting high-resolution or irregularly shaped visual assets.

키워드