本論文更將所提之靜態影像浮水印拓展至以場景為基礎之動態影片浮水印技術。藉由本論文所提之技術，在一段場景中經由每個畫面的浮水印萃取，最後終能取出完整之浮水印。在本論文中還探討了壓縮與共謀攻擊，實驗結果顯示，本論文所提以SVMs為基礎之浮水印技術是有效與具體可行。 Digital watermarking is an important technique for protection and identification of the content of digital multimedia. From our previous work on digital watermarking technique based on support vector machines (SVMs) and tolerable position map (TPM), a thorough study on feature selection and TPM generation is conducted. The improved system is endorsed by a complete set of experiments. It not only inherits the good characteristics from the original algorithm, but also shows significant improvements on time consumption as well as the robustness of low-pass filtering attacks. The proposed algorithm is also extended to a scene-based video watermarking technique. The excellent performance on NC values is achieved by averaging extracted watermarks from many frames of one scene into the final extracted watermark. The issue on collusion attacks is also detailed discussed. Our video watermarking technique is shown to be robust to video compression and collusion attacks. The proposed video watermarking technique has its novelty and practicability on SVM-applications.
