Digital watermark extraction technique for partially shadowed photos

This paper addresses the difficulty of detecting and/or verifying a watermark signal from watermarked image contained in unauthorized printing media. We consider a situation where an image owner tries to detect his/her embedded watermark from a photo having unauthorized image inside. This taken photo normally contains distortions such as those from the processes of printing and photography resulting in an inaccuracy of the extracted watermark. The situation becomes worse when the image to be detected is partially shadowed by some obstacles. We thus propose in this paper a method to efficiently detect a watermark from a partially shadowed photo. Essentially, an image registration technique base on projective transformation is used to diminish the RST (rotation, scaling, and translation) and perspective distortions from the printing and photography processes. Next, a shadow suppression technique based on gradient field is used to suppress unwanted noise-like shadow before detecting the watermark inside. The experimental results show effectiveness of our proposed method and possibility to implement it in other watermarking schemes.