Wavelet Transform-based SVD Watermarking Scheme

Abstract

In this paper, we propose a robust invisible digital watermarking scheme based on SVD on the wavelet transform domain. The first step of embedding process is wavelet decomposition of a given image up to a suitably chosen low scale. After partitioning the low scale image into 3x3 blocks, we choose watermark blocks by key sets: entropy and condition number of the block. The embedding process is finished with modifying the largest singular value to be watermark of each watermark block. As a result, this process is a robust watermarking in the lowest possible time-frequency domain. To detect the watermark, we are locally modeling an attack as a 3x3 matrix on each watermark block. By combining the SVD of the subimages with the attack matrices acted on the watermark blocks, we estimate watermark set corresponding to watermark blocks. The estimated watermark samples are justified as original watermark by the several statistical testing. Through numerical simulations, we show that the proposed watermarking scheme not only detects efficiently the watermarks but also is reliable from several attacks such as JPEG, Gaussian filtering, and sharpening.