Batch Steganography via Generative Network

Zhong, Nan; Qian, Zhenxing; Wang, Zichi; Zhang, Xinpeng; Li, Xiaolong

doi:10.1109/tcsvt.2020.2974884

Cited by 25 publications

(10 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To enhance steganographic security, Tang et al [41] present a steganographic scheme with adversarial embedding operation. Further, Zhong et al [42] propose a novel strategy of batch steganography exploiting a generative network. Through this method, different payloads can be allocated to different images.…”

Section: Steganographymentioning

confidence: 99%

HCISNet: Higher‐capacity invisible image steganographic network

Liu

et al. 2021

IET Image Processing

View full text Add to dashboard Cite

Information hiding as a crucial method for multimedia security and privacy protection, has received a substantial amount of attention. However, the most existing methods focus on the resistance of steganalysis and the robustness of information, resulting in low capacity. In this paper, to further increase the capacity of information hiding, a high-capacity adversarial image steganography model with end-to-end manner, termed as HCISNet is proposed. First, an enhanced Dense Atrous Spatial Pyramid Pooling module is presented to learn more semantic information in cover image, which can adaptively embed more message in redundant regions with rich textures. Then, a modified discriminator network with lightweight residual block is employed to assist the encoder network with optimal highcapacity solution. Meanwhile, the multiple objective function with the perceptual loss and several training tricks are developed, which can improve the visual and perceptual consistency of the original and the steganographic image, further enhancing the capacity. Finally, experiments and analysis are conducted on three public datasets, where this model has better imperceptibility, higher security and greater capacity. Under same experimental conditions, the cover image can embed 5.68 bits per pixel (BPP), far exceeding the previous highest value of 4.4 BPP in state-of-the-art methods in the authors' knowledge.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

show abstract

Section: Steganographymentioning

confidence: 99%

HCISNet: Higher‐capacity invisible image steganographic network

Liu

et al. 2021

IET Image Processing

View full text Add to dashboard Cite

show abstract

“…Image steganography Inspired by the point of Goodfellow et al that "adversarial examples can be thought of as a sort of accidental steganography" [5], we treat adversarial perturbations as special steganographic modifications from the view of steganography [11]. These special steganographic modifications are added in original images to fool classifiers and avoid detections.…”

Section: Related Workmentioning

confidence: 99%

“…C&W minimizes the adversarial loss and the distance loss simultaneously to achieve the attack and keep the visual quality. We add the microscopical regularization to the object loss function of C&W attack as (11).…”

Section: Combination With Candwmentioning

confidence: 99%

Undetectable Adversarial Examples Based on Microscopical Regularization

Zhong

Qian

Zhang

2021

2021 IEEE International Conference on Multimedia and Expo (ICME)

Self Cite

View full text Add to dashboard Cite

Recent works have demonstrated that neural networks are vulnerable to adversarial examples. Although existing methods have achieved satisfactory attack success rates, most adversarial examples can be detected by statistical analysis and further removed. In previous methods, adversarial perturbations are added using adversarial loss and distance metrics, in which the positions of modified pixels are not considered. In this paper, we elaborate a microscopical regularization that introduces adversarial perturbations onto rich texture regions. The microscopical regularization is used to evaluate pixellevel differences between a normal image and its adversarial version. We further propose a novel optimization strategy of modification probability matrices to minimize the loss function that satisfies the restriction of L ∞ . Through extensive experiments, we show that our method can resist statistical analysis by a large margin and achieve better visual quality than others. The proposed microscopical regularization can also be combined with existing approaches to enhance the undetectability and robustness.

show abstract

“…For the receiver, secret message can be extracted by detecting the encoded points in the constructed fingerprint. In [44], batch steganography using a generative network is proposed to information in multiple images. In order to counter steganalysis by trying to fool the machine learning classifiers, researchers in [45] proposed an adversarial embedding technique to generate adversarial stego images with minimum amount of adjustable elements.…”

Section: Fig 2 General Model Of Information Extraction and Original Media Recoverymentioning

confidence: 99%

“…Recently, some researchers have applied neural networks to steganography [35][36][37][38][39][40][41][42][43][44][45]. As a cooperative algorithm, convolutional neural network with deep supervision edge detector retains more edge pixels over conventional edge detection, so it can increase data hiding capacity [35].…”

Section: Fig 1 a Simple Model Of Information Hidingmentioning

confidence: 99%

Large-Capacity Image Data Hiding based on Table Look-up

Ding¹,

Zhang²,

Reulke³

et al. 2021

Preprint

View full text Add to dashboard Cite

In previous data hiding techniques, binary rules are usually used to guide the fine-tuning of the values of basic objects in the host media to hide bit 0 and bit 1. In this paper, we propose a new data hiding technique for gray images based on querying a 256x256 information table. The information table is constructed by cloning a 3x3 basic block, which we call seed block. Eight unsigned integer values between 0 and 7, i.e., 3 bit binary data, are assigned to different elements of the seed block. Each time, a pair of pixels are chosen from a host image, and their pixel values are used as row and column numbers to look up the information table. If element value obtained by looking up the table is equal to the 3 bit binary data to be hidden, the values of the pixel pair will remain unchanged. Otherwise, take this element as the central point, we call it the focus element, to enclose a 3x3 window in the information table. Then in the window, find the element which is equal to the data to be hidden. Finally, update the pixel values of the pair with the row and column numbers of the found element in the window. Since the row and column numbers are in the range of 0-255, the updated pixel values will not overflow. In the proposed algorithm, a pair of pixels can hide 3 bits of information, so the embedding capacity is very high. Since the adjustment of pixel values is constrained in a 3x3 window, the modification amount of pixel values is small. The proposed technique belongs to fragile digital watermarking, so it can be used for image authentication and tamper localization. By the evaluation of data hiding capacity, security, imperceptibility, computational cost and extensibility, this algorithm is superior to existing information hiding techniques. The proposed technique can also be used in color image and audio data hiding.

show abstract

Batch Steganography via Generative Network

Cited by 25 publications

References 23 publications

HCISNet: Higher‐capacity invisible image steganographic network

HCISNet: Higher‐capacity invisible image steganographic network

Undetectable Adversarial Examples Based on Microscopical Regularization

Large-Capacity Image Data Hiding based on Table Look-up

Contact Info

Product

Resources

About