【作者】 王博；

【导师】 王丽娜；

【作者基本信息】 武汉大学 ， 网络空间安全， 2019， 硕士

【摘要】 隐写术是信息隐藏领域的热门研究方向,通过该技术可以将隐秘信息嵌入到公开多媒体文件中且不引起怀疑,从而保证通信安全。近年来,自适应隐写方法的提出,显著地提高了隐写术的安全性。如何进一步提高隐写术的安全性和适用性是隐写术领域的关键问题。目前用于隐写的图像格式主要包括JPEG图像和空域图像两种。基于JPEG图像的隐写方法通常利用原始图像的边信息来提高隐写安全性。然而现有的JPEG隐写术在使用边信息时需要提前获取原始图像,当原始图像不可获得时,往往不能把边信息用于设计隐写方法。而空域图像缺少类似于JPEG图像的压缩过程,所以关于利用边信息的空域图像隐写术的研究较少。本文以JPEG图像和空域图像作为隐写载体,利用自适应隐写框架,提出了基于边信息估计的图像隐写方法,并对其安全性开展了相关研究,取得的研究成果如下:1)基于边信息估计的JPEG图像隐写方法对于目前的JPEG图像隐写术,在没有原始图像的情况下,大多数方法往往不能利用边信息来辅助隐写嵌入。本文提出了一种利用卷积神经网络估计JPEG边信息的方法,并利用估计的边信息优化嵌入修改操作。并且将该方案应用到两种现有的JPEG自适应隐写方法JC-UED和J-UNIWARD上,得到两种新的隐写方法。实验结果表明,该方案在只能获取JPEG载体图像的情况下,利用估计的边信息提高了原隐写方法的安全性。2)利用代价调整因子的JPEG图像隐写方法在基于边信息估计的JPEG隐写方法的基础上,引入代价调整因子,进一步提高该方法的安全性。在获得DCT系数的估计边信息后,不直接确定修改嵌入的方向,而是用调整因子对相应修改方向的失真代价值进行调整。通过该方法可以进一步充分利用估计的边信息,同时应用三元STC编码进行嵌入,进一步提高了隐写的安全性。实验结果表明,在没有原始图像的情况下,本方法相比其他两种隐写方法具有最高的抗检测能力,而且在高嵌入率的情况下,也能保持较高的抗检测能力。3)基于边信息估计的空域图像隐写方法在空域图像隐写方面,将基于边信息估计的思想应用到空域图像隐写中。空域图像的边信息通常是由源图像进行信息量减少的变换过程中得到的。而在不具有源图像的情况下,本文方法利用一种超分辨率方法SR-CNN对载体图像进行超分辨率计算,然后通过下采样算法计算得到估计的边信息。利用估计的边信息和代价调整因子对像素的代价值进行调整,从而改进原有的空域隐写方法。实验结果表明,基于边信息估计的空域图像隐写方法可以在一定程度上提高原隐写方法的安全性能。本文围绕基于边信息估计的图像隐写方法进行研究,以JPEG图像和空域图像为载体,提出了基于边信息估计的JPEG图像隐写方法,在此基础上提出了利用代价调整因子的JPEG图像隐写方法和基于边信息估计的空域图像隐写方法,提高了原有隐写方法的安全性能。更多还原

【Abstract】 Steganography is a hot research direction in the field of information hiding.This technology can embed secret information into public multimedia files without causing suspicion,thus ensuring communication security.In recent years,the introduction of adaptive steganography has significantly improved the security of steganography.How to further improve the security and applicability of steganography is a key issue in the field of steganography.The image formats currently used for steganography mainly include JPEG images and spatial domain images.Steganography methods based on JPEG images typically utilize side information of the original image to improve the security of steganography.However,the existing JPEG steganography methods need to obtain the original images in advance when using the side information.When the original images are not available,the side information cannot be used to design the steganography methods.While spatial images lack a compression process similar to JPEG images,there is less research on spatial image steganography using side information.In this paper,JPEG image and spatial domain image are used as steganographic covers.With adaptive steganography framework,image steganography based on side information estimation is proposed.The security research about this steganography framework is carried out.The research results are as follows:1)JPEG image steganography based on side information estimationFor current JPEG image steganography,most methods cannot use side information to help embed in the absence of the original image.This paper proposes a method for estimating JPEG side information using convolutional neural networks,and then uses the estimated side information to optimize the embedding modification.And the scheme is applied to two existing JPEG adaptive steganography methods JC-UED and J-UNIWARD,and two new steganography methods are obtained.The experimental results show that the scheme improves the security of the original steganography methods by using the estimated side information when only the JPEG cover image can be obtained.2)JPEG image steganography using cost modulation factorBased on the JPEG steganography method based on side information estimation,the cost modulation factor is introduced to further improve the security of steganography.After obtaining the estimated side information of the DCT coefficients,the directions of the modification are not directly determined,but the modulation factor is used to modulate the distortion costs of the corresponding modification direction.The method can further make full use of the estimated side information,and simultaneously apply the ternary STC code for embedding,thereby further improving the security of steganography.The experimental results show that this method has the highest antidetection ability compared with the other two steganography methods without the original image,and can maintain high anti-detection ability under the condition of high payloads.3)Spatial image steganography based on side information estimationIn the aspect of spatial image steganography,the idea based on side information estimation is applied to spatial image steganography.The side information of the spatial image is usually obtained from the source image using some information-reducing operation.Without the source image,the proposed method uses a super-resolution method SR-CNN to perform super-resolution calculation on the cover image,and then calculates the estimated side information through the down-sampling algorithm.The original spatial steganography method is improved by using the estimated side information and the cost modulation factor to modulate the costs of pixels.The experimental results show that the spatial image steganography method based on side information estimation can improve the security performance of the original steganography method to some extent.In this paper,the image steganography method based on side information estimation is studied.Using JPEG image and spatial domain image as covers,a JPEG image steganography method based on side information estimation is proposed.Based on this method,JPEG image steganography method using cost modulation factor and Spatial image steganography method based on side information estimation are proposed.The security performance of the original steganography methods is improved.更多还原