Research On Tensor-based Video Watermarking Algorithms

With the rapid development of the Internet,video has gradually become one of the main ways of information transmission.At the same time,due to the endless stream of video piracy,the need for video copyright protection has become increasingly urgent.Among them,video watermarking technology is one of the most effective methods to achieve video copyright protection.However,most of the current video watermarking algorithms perform frequency domain transformation on a single frame of video,which results in the ability of the algorithm to resist frame attacks needs to be improved.Since the tensor can intuitively represent the structural characteristics of high-dimensional data,representing a video as a tensor can maximize the correlation on the video time axis to improve the robustness of the algorithm to frame attacks.In this paper,by performing Tucker decomposition on the video tensor and studying the core tensor and factor matrix after decomposition,three video watermarking algorithms based on tensor are proposed.The main work and innovations of this article are as follows:(1)Aiming at the problem that most current video watermarking algorithms perform frequency domain transformation on a single frame of video,which results in the poor ability of the algorithm to resist frame attacks,this paper proposes a video watermarking algorithm based on core tensor.This algorithm obtains a stable core tensor and three factor matrices by performing Tucker decomposition on the video tensor.Since the core tensor represents the main energy of each frame of the video tensor,the robustness of the algorithm can be further improved by embedding watermark in the core tensor.Moreover,the core tensor embedded with the watermark is uniformly dispersed into each frame of the video tensor through the Tucker decomposition inverse transform.The experimental results show that the core tensor-based video watermarking algorithm has good transparency and is robust to attacks such as frame deletion,video compression,rotation,scaling,noise addition,and filtering.(2)Aiming at the problem that strong robust video watermarking algorithms often cause the image quality of video frames to significantly decrease,this paper proposes a video watermarking algorithm based on tensor feature maps.In the watermark embedding stage,the video tensor and the transposition of the factor matrix containing the time axis information are modulo multiplied to obtain the large feature tensor such as the video tensor.The first forward slice of the feature tensor contains the main information of each frame of the video tensor,which is called the tensor feature map.This algorithm embeds watermark on the tensor feature map,so that the watermark information is evenly embedded in each frame of the video tensor,which further improves the transparency of the algorithm.The experimental results show that the video watermarking algorithm based on tensor feature maps has higher transparency and is more robust to frame attacks.(3)Aiming at the problem that the robustness of current video watermarking algorithms to geometric deformation and cropping attacks still need to be improved,this paper proposes a video watermarking algorithm based on time factor matrix.This algorithm obtains the core tensor and three factor matrices by Tucker decomposition of the video tensor.The three factor matrices represent the row information,column information,and time axis information of the video frame,respectively.The algorithm calls the factor matrix containing time axis information as the time factor matrix.The value in the first column of the time factor matrix represents the main correlation coefficient of each frame of the video tensor.Due to geometric deformation or cropping attacks on the video,the correlation coefficient between adjacent frames of the video does not change significantly.Therefore,the algorithm embeds the watermark information by modifying the values of the first column of the time factor matrix,thereby further improving the robustness of the algorithm against geometric deformation and cropping attacks.The experimental results show that the video watermarking algorithm based on the time factor matrix is not only more robust against cropping,scaling,rotation and other attacks,but also has strong resistance to video-specific attacks such as frame deletion and video compression.