Noise Simulation-Based Deep Optical Watermarking

Digital watermarking is an important branch of information hiding, which effectively guarantees the robustness of embedded watermarks in distorted channels. To embed the watermark into the host carrier, traditional watermarking schemes often require the modification of the carrier. However, in some cases, the modification of the carrier is not allowed such as paintings in museums. To address such limitation, we utilize optical watermarking to embed the watermark into the host carrier. Optical watermarking refers to a technique that encodes the watermark into the visible light irradiating the object, where the watermark can be further extracted by the camera photography process. To realize transparency and robustness of the watermark, we propose a color-decomposition-based watermarking pattern generation algorithm which satisfies human visual system (HVS) characteristics, a camera shooting simulation algorithm which accurately produces the dataset for training, and a decoding network which can realize loss-less decoding of the embedded watermark. Various experiments demonstrate the superiority of our method and reveal the broad applicability of the proposed technique.