最小依赖隐藏的屏摄鲁棒水印方法

LDH:least dependent hiding for screen-shooting resilient watermarking

目的 现有屏摄水印方法无法有效平衡计算复杂度、嵌入水印后的图像质量以及水印鲁棒性3项指标,同时广泛使用透视畸变矫正预处理,大大限制了屏摄水印的实际商业使用。本文在重新设计噪声层的基础上,提出了一种最小依赖载体图像隐藏水印信息的屏摄鲁棒水印,将屏摄水印对于载体图像的依赖控制在最小。方法 为了保证水印的嵌入效率,极大简化依赖深度隐藏网络框架中的编码网络,达成对载体图像的最小依赖,大大减小计算复杂度;为了平衡网络深度减小所导致的网络提取能力损失,加入Sobel算子,引入载体图像的边缘信息;在噪声层中加入缩放攻击操作,并由此去除了限制屏摄水印应用范围的透视畸变矫正预处理,进一步拓宽了应用范围;为了训练网络的屏摄鲁棒性,重新定义了噪声层,改进原有噪声层的设计结构,对噪声层图像扰动类型和参数进行随机选择,使得解码网络的输入数据具有更高的样本均衡性和多样性。结果 在DIV2K(DIVerse 2K)数据集上与其他的3种方法进行了对比实验,本文方法获得了最高的PSNR(peak signal-to-noise ratio)和SSIM(structural similarity index measure)指标,并比排名第2的通用深度隐藏方法提高了12 dB的PSNR值和0.006的SSIM值;在有无攻击两种环境下,本文方法均能保持很高的ACC(accuracy)和F1指标,在攻击环境下比排名第2位的StegaStamp(steganography stamp)方法提高了0.262的F1分数。与同网络框架下的已有噪声层相比,在无攻击环境下,本文算法提高了0.124的ACC和0.284的F1分数;在有攻击环境下,本文算法提高了0.316的ACC和0.524的F1分数,水印提取的准确性更高。结论 本文算法在图像质量和水印鲁棒性方面获得了更优的效果,摆脱了透视畸变矫正的限制,拓宽了屏摄水印的应用范围。

Objective With the rapid development of internet and communication technology,the remote desktop techniqueenables separating the confidential information and the screen in space.However,it also engenders information securityrisks of confidential information because of illegal screen shooting.How can illegal screen shooting be prevented and the related responsibility identified?Adding a robust watermark and revealing the message hidden in the shot image is pre⁃ferred.By taking photos of the files displayed on the screen,the captured photos can realize efficient,high-quality informa⁃tion recording.The pictures taken on the screen not only record effective information but also destroy the possible water⁃mark signal carried to a large extent,making the photo leakage behavior concealed and difficult to trace.Screen-shootingwatermark is a challenging subject in digital watermark.In screen shooting,the information displayed on the screen isreceived through camera capturing and postprocessing operations to transmit information from the screen to the camera inthe optical channel involving optical changes,digital-analog conversion,image scaling,and image distortion.Four mainmethods are used to deal with this subject,namely,key-point-,template-,frequency-domain-,and deep neural network(DNN)-based methods.Traditional methods and DNN-based methods have some solutions.However,neither of themcould balance computational complexity,image quality,and watermark robustness.The calculation of key points in keypoint based methods is always overly time-consuming for practical use.Template-based methods often bring great changesto the cover images,resulting in image quality degradation.Watermarks generated by the frequency-domain-based methodshave poor robustness and could be easily destroyed.Almost all methods should correct and resize the warped image to itsoriginal image size for the following watermark extraction stage,which is the main reason why the watermarks in these meth⁃ods could not achieve robustness to clipping and scaling in practice.To solve the above problems,the least dependent hid⁃ing for screen-shooting resilient watermarking method is proposed to consider computational complexity,image quality,and robustness comprehensively.The decoder-based reveal network only needs to disclose the watermark message from thecorresponding location of the container image,which guarantees the semantic consistency of the reveal network and theembedding network.The embedded watermark,such as user name,time,and IP address,could be extracted under thescreen-shooting attack or other attacks,and to imitate the information loss in screen shooting,an improved noise layer isdesigned for the training of our model.Method First,the watermark embedding network in the dependent deep hiding(DDH)framework is greatly simplified,and the Sobel operator is added to introduce the edge information of the coverimage.The scaling attack operation is added to the noise layer,and the perspective distortion correction preprocessing isremoved because it limits the application range of screen-shooting resilient watermarking.The existing noise layer is rede⁃fined in the way that the image disturbance types are randomly selected and the parameters of the specific image distur⁃bance types are randomly changed,which increases the sample equilibrium and diversity of the training data of the revealnetwork.The investigation of previous DNN-based methods reveals their watermark residuals visually approximate theedges of the cover images.A strong correlation exists between the edges of the cover images and the invisibility of the water⁃mark.To improve robustness and reduce computation complexity,the edge map of the cover image extracted by the Sobeloperator is concatenated with the feature map of the watermark.The watermark embedding network is divided into two partsaccording to whether the cover image is used in the convolution because the network part without cover image participatingin it could be previously calculated in practice.Second,the existing noise layer is modified to simulate the image scalingoperation in the screen shooting,so the widely used perspective distortion correction can be canceled.Considering theclass-balance principle,a new design idea of noise layer is proposed,in which random decision modules are added to thenoise layer to make the data augmentation stronger than the original image disturbing effects.When training the network,learned perceptual image patch similarity(LPIPS)loss,L2 loss,and structural similarity index measure(SSIM)loss areused to constrain the visual similarity of the cover image and the container image while information entropy loss andweighted cross entropy loss are used to reconstruct the watermark with the form of a single-channel binary image.Modeltraining and testing is carried out based on PyTorch.PyTorch is used to implement least dependent hiding(LDH)withNVIDIA GeForce 2080Ti GPU and Intel Core i7-97003.00 GHz CPU.The whole neural network is optimized by Adamoptimizer.The initial learning rate is set to 1e-3,which is then reduced 90%every 20 epochs.In the training,the inputimage resolution is 256×256 and the batch size is 2.A pretrained model trained without geometric transformation in thenoise layer is used to initialize the model.Result Experimental results show the proposed noise layer is more effective thanthe three latest methods on the DIVerse 2K(DIV2K)dataset.The proposed method achieves the highest peak signal-tonoise ratio(PSNR)and SSIM index,which improves PSNR by 12 dB and SSIM by 0.006 compared with the second-bestmethod—universal deep hiding(UDH)if no image attacks are applied.Moreover,it ranks second in accuracy and F1 index if no image attacks are applied.Compared with the same network framework using the noise layer proposed by theprevious work,our algorithm achieves better indicators and higher accuracy for the watermark extraction in both modes withand without image attacks,which proves the noise layer proposed is indeed helpful to increase the training to improve theaccuracy and robustness of watermark extraction.The watermark can be extracted from the screen shot images in the rangeof 10 cm to more than 50 cm,and it has a high extraction success rate at a usual distance.Conclusion In this paper,theleast dependent hiding for screen-shooting resilient watermarking is proposed,which comprehensively balances computa⁃tional complexity,image quality,and robustness.An effective noise layer improvement measure is also designed,whichhelps our algorithm perform better in image quality and watermark robustness.The proposed algorithm has the advantagesof high embedding efficiency,high robustness,and high transparency,which means wider application range comparedwith the existing methods.