Current Issue Cover

贾孟霖, 杨杨, 孙冬(安徽大学)

摘 要
目的:图像隐藏已成为计算机视觉领域的一个重要课题,其目的是以难以察觉的方式将秘密图像隐藏在载体图像中,同时要求接收端能够恢复秘密图像。尽管该技术发展迅速,但目前的图像隐藏技术大多是从内容层面进行伪装,追求载密图像与载体图像的不可区分性。其实,图像隐藏的本质是对行为安全的追求,因此不仅可以在内容层面进行伪装,还可以在行为层面进行伪装。 方法:本文从行为安全的角度出发,提出了一种基于超分辨率行为伪装的可逆图像隐藏方法。与传统的图像隐藏技术不同,本文首先将秘密图像可逆地隐藏到载体图像中,生成载密图像,然后通过可逆的超分辨率处理创建与普通超分辨率图像处理操作无法区分的伪装图像。最后,允许接收方从伪装图像中恢复秘密图像和载体图像。 结果:在图像隐藏和超分辨率两个任务中,我们的方法均取得了优异的结果。在相同的数据集下,测试结果显示恢复秘密图像的峰值信噪比(peak signal-to-noise ratio, PSNR)值达到47+dB,较其他方法提升了2%以上,结构相似度(structure similarity index measure, SSIM)值也达到了0.99+,超分辨率图像的峰值信噪比(PSNR)提升了2+dB, 感知指数(perceptual index, PI)值降低了2.02+。 结论:本文所提出的图像隐藏框架利用可逆超分辨率处理操作实现了行为安全角度的图像隐藏,在容量、安全性、精度上都具有优势。
Reversible image hiding method based on super-resolution behavior camouflage

Jia Meng Lin, Yang Yang, Sun Dong(Anhui University)

Objective:Image hiding is recently a hotspot of computer vision, which aims to hide the secret image into the cover image in an imperceptible way, and then recover the secret image as much as possible at the receiver. Traditional image hiding methods often adjust the cover image"s pixel value in the spatial domain or modify the cover image"s frequency coefficients to hide the secret information. Since these methods hide secret images through handcrafted feature information, they are easy to be detected by existing detection techniques. Apart from their weak security, these methods lack significant capabilities for hiding information in images and therefore fall short of meeting the demands of large capacity image hiding tasks. With the advancements in convolutional neural networks, deep-learning-based image hiding has been quickly developed. These deep learning methods seek to achieve a high level of capacity, invisibility, and recovery accuracy. However, with the rapid development of steganalysis, existing image hiding techniques are easily detected by deep learning analysis methods. Whether handcrafted image hiding methods or deep-learning-based image hiding methods, they are all camouflaged from the content level, pursuing the indistinguishability of the marked image and the cover image. In fact, the essence of image hiding is the pursuit of behavioral security, that is, the pursuit of hiding secret information is inseparable from the behavior of normal users, so as to achieve better detection resistance. Therefore, we can not only camouflage at the content level but also disguise at the behavior level. Therefore, from the perspective of behavior security, we innovatively use super-resolution, a common image processing technology, as our behavior camouflage means, so as to realize image hiding. Method:In general, traditional image hiding techniques tend to prioritize the indistinguishability of the cover image and the secret image in the content level, but in this paper, we aim to achieve image hiding from a behavioral security perspective. Specifically, we aim to make steganographic behavior indistinguishable from regular super-resolution image processing behavior. The entire method can be divided into three modules: forward hiding, super-resolution behavior camouflage, and backward revealing. In the first module, a cover image and a secret image are inputted to the forward hiding module, resulting in a marked image that looks identical to the cover image but with hidden information and information that is lost during forward hiding. The second module involves a lightweight super-resolution rescaling network to realize behavior camouflage. Instead of using traditional convolution for up and down sampling, we use a bicubic interpolation. At the same time, we use a pre-trained VGG19 network to extract high-level features to guide the generation of super-resolution behavior camouflage images. The final module is backward revealing, where the marked image is first reconstructed using the reversibility of the behavior camouflage module. The reconstructed image and auxiliary matrix are then inputted into the backward revealing module to recover the secret image and the cover image. Result:Experiments were carried out on recovering secret images, camouflage effect of super-resolution behavior, parameter setting, ablation experiment, and so on. The results show that our invisibility, hiding capacity, recovery accuracy, and so on have reached a good level, only the recovery accuracy is slightly lower than the current SOTA. Specifically, the peak signal-to-noise ratio (PSNR) between the secret image and the recovered secret image can reach 47.23dB which is about 0.92dB less than SOTA, and the structural similarity (SSIM) can reach 0.9938 which is 0.0034 less than SOTA. But compared with other methods besides SOTA, our method has obvious advantages. At the same time, our super-resolution behavior camouflage has reached a satisfactory level. Due to reversibility constraints, we are not looking for top-of-the-line super-resolution effects. Our super-resolution behavior camouflage images have a PSNR of 27.43dB and a Perceptual Index(PI) of 4.5684. Whether subjective or objective, it has reached the satisfaction of human eyes. In addition to the above two main indicators, we also conducted exploratory experiments on the selection of super parameters, module architecture, and loss function in some experiments to find the optimal setting so as to achieve a better combination effect. Conclusion:This paper proposes a new idea of image hiding in which in the process of hiding the secret image, super-resolution processing is carried out at the same time to obtain a super-resolution behavioral camouflage image with secret information, so as to divert the attention of unauthorized parties and realize the protection of the secret image. The experimental results show that our method can not only achieve high capacity, high invisibility, and high recovery accuracy but also play a good role in the confusion of unauthorized parties, camouflage images still maintain good visual effects.