| In recent years, Internet technology develops rapidly, getting information from Internet is becoming convenient and efficient. Therefore, Internet has become the main approach to get information. With the rapid growth of digital multimedia technology, multimedia information is becoming easy to be produced and broadcast. This promotes the use of multimedia, but at the same time, it also brings serious copyright issues. The digital watermark technology is a new mean for copyright protection.Digital watermark is a technology that refers to multimedia as a carrier, and embeds identification information (such as copyright information, etc.) into the carrier. Digital watermark technology should meet the robustness, transparency, security and so on. In this paper, aiming at poor robust and transparent performance of spatial image watermark technology, we propose and implement two algorithms-an adaptive watermark embedding strength calculating algorithm based on machine learning and an spatial digital image watermark algorithm based on machine learning. The adaptive watermark embedding strength calculating algorithm in this paper calculates the maximum watermark embedding strength that satisfies transparent performance adaptively by training machine learning model. The trained machine learning model learns the relationships between image block, watermark embedding strength and transparency. As the experimental results show, the proposed algorithm increases the embedding strength greatly with small visual performance influence. The spatial digital image watermark algorithm based on machine learning learns the relationships between pixel and its neighbour pixels. The proposed algorithm embeds and extracts watermark information with the relationships between the outputs of the machine learning model and image pixels. As the experimental results show, compared with current spatial watermark algorithm, the proposed algorithm improves the robust and transparent performance. |
