The Key Technology Research Of Digital Video Watermark Based On Machine Learning

This paper mainly researches the video watermark technology in machine learning , including the support vector machines and integrated learning. First classifies the new kind of the video watermark attack, then gives the prevention methods. Because the video is composed of multi frame, different frames are embedded the watermark by different algorithms, we designed various of algorithms, in order to improve the single algorithm robustness defects (single algorithm could not resist all attacks).Based on the analysis,the paper then mainly analyzes the most vulnerable attack, collusion attack. The preventions against collusion attack at home and abroad are summarized, typical algorithms are given, for the future of collusion attack thoughts. then provides new directions, to lay a foundation of overcoming this attack.According to the characteristics of the video and texture features , a new video watermarking algorithm is proposed based on the HVS, making invisibility and robust achieve the best compromise and laying the foundation of the support vector machine application in the digital watermark.Support vector machine is an effective method kind of the machine learning , mainly used for the classification problems, but also can be used for the regression problems. Because support vector machines and digital watermarking exist effective conjunction, many scholars applied support vector machine to the digital watermarking, but now support vector machine is generally applied to image and audio, the application of the video is too few, this paper provided new thoughts about how we apply the SVM to the video, based on the image watermark technology.