An automatic visible watermark detection method using total variation

Visible watermarking has been an important research area, widely studied and used for copyright protection over the years. However, new visible watermark removal-oriented attacks and techniques have emerged. Although these methods provide high quality resultant images, the main part of the attack consists of manual detection of the watermark pattern. Considering that visible watermark pattern can be different in each watermarked image, time-consuming human intervention is required, making these attacks inefficient in practice. In this paper, we propose an automatic detection method of visible watermark pattern, which can combine with conventional watermark removal attacks to make them more efficient. To accomplish it, the visible watermarked image is decomposed into structure image and texture image using Total Variation L1 (TV-L1) method. In the structure image, the watermark edge is discriminated from the edges present in the host image to automatically detect the visible watermark pattern.