A video zero-watermarking algorithm based on text detection

The video text contains important information of the video content, such as the Chinese oral-history videos which record some dying craft through interviewing with some folk artists. This paper proposes a zero-watermarking algorithm based on text detection for copyright protection and content authentication, which includes a unified framework for text detection, text segmentation and video zero-watermarking scheme. Firstly, according to the characteristics of Chinese strokes, four directions of first-order differential filter are adopted for filtering video frames, and the texture images without most background can be acquired. Secondly, for generating the horizontal long rectangle text regions , the morphologic `close´ and `open´ operations are employed to deal with the texture images. Thirdly, a text segmentation method based on improved twice time OTUS[1] is used to segment the text regions, then the final zero-watermarking information is generated and preserved in the third-party institutions after normalizing text regions. Finally, the zero-watermarking information is extracted from the unverified video, and matched with the original one. Experimental results show that the proposed zero-watermarking algorithm has good robustness for the video attacks such as frame rate changing, recoding, frame dropping, transcoding and tampering text.