Research On Quantization Watermarking Based On Robust Visual Properties

Nowadays,the popularity of smart terminals,such as mobile phones and personal computers,not only provides convenience for people's lives,but also brings a series of problems.On the one hand,various image and video editing tools on the terminal make it easier to tamper with the original data,which leads to a series of piracy problems;on the other hand,with the diversification of multimedia data content,the security protection and authentication for different types of multimedia data become more difficult.Among them,as an important technical means of data security protection and authentication,digital watermarking technology has developed rapidly in the past decade,and has played an important role in solving security problems such as copyright protection and bill anticounterfeiting.However,in the current environment of explosive growth and security problems,how to guarantee and improve the performance of watermarking algorithm has always inspired researchers to move forward.As two important technical indicators of digital watermarking technology,robustness and invisibility,the contradiction between them limits the further application of the algorithm,so how to achieve the performance of both at the same time is always the core problem of robust watermarking algorithm design.In the traditional solutions,the robust watermarking algorithm based on the visual characteristics has attracted researchers' attention because of its achievements in visual fidelity and algorithm robustness.However,in the actual feature design process,many algorithms cannot guarantee the consistency of the visual reference features in the watermark embedding and detection end,which will lead to the performance degradation of the watermark detection.In addition,with the development of human visual system research and visual computing,more and more visual features appear to represent the characteristics of human visual system.Thus,this paper fully studies and utilizes the visual characteristics of human vision system in the process of image perception,designs the corresponding robust visual features,reasonably distributes and guides the watermark processing,and realizes the effective improvement of robustness and invisibility.Specifically,the work of this paper is carried out from the following aspects:First of all,in this paper,we first propose a watermarking algorithm guided by the just noticeable distortion(JND)model based on pattern complexity,which considers the masking effect of different image pattern complexity on the visual content and uses three alternating current(AC)coefficients of image block after discrete cosine transform(DCT)to effectively represent the energy in different directions.The pattern complexity is calculated based on the energy of each direction and defined as a new visual masking factor to determine the value of JND model.And then,the quantization strength of each image block is calculated in the framework based on the proposed JND model to improve the invisibility of the watermark information and the robustness of the algorithm.Secondly,based on the human visual system's perception of image orientation and color attributes,this paper proposes a color image watermarking algorithm based on directional diversity and color complexity measures.In this scheme,a new direction feature extraction method is used to calculate the direction feature of each image block in the image luminance channel,and the visual masking effect of the surrounding image blocks on the current block is analyzed by the direction feature.In addition,in order to reflect the diversity of color difference perception caused by color change in human vision during the process of color image observation,the color complexity masking effect of each image block is calculated by using the color information of image chroma channel.Finally,we combine two new visual masking features to get the robust perceptual JND model.Through the proposed visual JND model,we design watermark embedding and extraction,and realize the effective trade-off between image quality and algorithm robustness in watermark processing.Thirdly,by using the difference of attention distribution in different regions of the image by human brain to represent the visual sensitivity,this paper proposes a JND model watermarking algorithm based on visual salience(VS)guidance.Firstly,the focus value of the image is calculated in DCT domain and the Top-Down salient features are constructed according to the focus region.Then,the visual attention map is mapped to the weighted map by using the central concave theory,and a new robust VS model of two-layer fusion is obtained by combining the traditional brightness and texture features.By using the VS model to modulate JND model and guide watermark processing,the image quality and robustness of the algorithm can be improved.Fourth,for special screen content images(SCIs),this paper proposes a hybrid JND model guided watermarking algorithm for color screen content images.The algorithm considers the difference between text content and image content in SCIs,classifies text blocks and image blocks by using the statistical difference between the two contents,and designs the corresponding visual JND model to guide the watermark processing according to the visual masking difference presented by different block types,so as to achieve the balance between the watermark image quality and algorithm robustness.Finally,in order to solve the problems of the complexity of feature design and the difficulty of embedding criteria selection in the design process of traditional watermarking algorithm,this paper proposes a new robust digital watermarking technology based on deep learning framework.The algorithm uses the convolutional neural network(CNN)realizes multi-level visual feature extraction,and uses watermark detection and loss function before and after embedding to guide watermark embedding.Through the continuous iteration of embedding and extraction process,the model is updated and optimized continuously.The optimal embedding position and embedding size are found,and finally the watermark image with high quality and stronger algorithm robustness are obtained.