Fig 1 - uploaded by Ta Minh Thanh
Content may be subject to copyright.
The architecture of our image steganography system. The two components of the full system. Left: Hiding the secret image in the cover image with Unet. Right: Revival the hidden image with the revival CNN network.
Contexts in source publication
Context 1
... original U-net is mainly applied to biomedical image segmentation; however, we find that the architecture of U-net is suitable for hiding network. The overall architecture of our network is demonstrated as in Figure 1. Besides, a new training scheme is also proposed. ...
Context 2
... networks are responsible to encode the secret image into the container image and decode the container image to get the original secret image. The whole processing pipeline is shown in Figure 1. This structure is similar to the architecture that appear in auto encoder [20], generative adversarial networks (GAN) [11], etc. ...
Similar publications
Early Alzheimer's Detection and Classification using VGG Convolutional Neural Network and Systematic Data Augmentation using MR Images
![Fig. 1. Architecture of LeNet5 [10].](profile/Ida-Nurhaida-2/publication/359821539/figure/fig1/AS:1148769030676480@1650899288127/Architecture-of-LeNet5-10_Q320.jpg)
Citations
... However, these traditional methods of digital steganography, though effective in hiding data, can potentially suffer from limitations such as limited data embedding capacity, vulnerability to image compression and noise, and trade-offs between data security and computational complexity (9,17,18). Alternative techniques, such as adaptive steganography (19)(20)(21) and deep learning-based steganography (22)(23)(24)(25)(26) offer more secure solutions, which, however, can be computationally intensive and demand active digital computing resources for both information hiding and recovery. In general, there is a growing need for alternative approaches that combine high speed, energy efficiency, and robustness and versatility for private information concealment. ...
We introduce an information-hiding camera integrated with an electronic decoder that is jointly optimized through deep learning. This system uses a diffractive optical processor, which transforms and hides input images into ordinary-looking patterns that deceive/mislead observers. This information-hiding transformation is valid for infinitely many combinations of secret messages, transformed into ordinary-looking output images through passive light-matter interactions within the diffractive processor. By processing these output patterns, an electronic decoder network accurately reconstructs the original information hidden within the deceptive output. We demonstrated our approach by designing information-hiding diffractive cameras operating under various lighting conditions and noise levels, showing their robustness. We further extended this framework to multispectral operation, allowing the concealment and decoding of multiple images at different wavelengths, performed simultaneously. The feasibility of our framework was also validated experimentally using terahertz radiation. This optical encoder–electronic decoder-based codesign provides a high speed and energy efficient information-hiding camera, offering a powerful solution for visual information security.
... So far, there is no published deep learning-based HDR image steganography algorithm and also no algorithms of "hiding image in HDR image," we compared the proposed method with deep learning-based algorithms of "hiding image in LDR image" [5,10,26,44,45]. Figure 5 shows the cover and stego images of all the compared algorithms. From the first row to the last row are the HDR cover images (tone mapped version of HDR cover images of ours, thus, also LDR cover images of the algorithms considered in comparison), stego images of our method, Chen et al.'s method [44], Van [45] and our method are almost as same as that of the cover images. ...
... Figure 5 shows the cover and stego images of all the compared algorithms. From the first row to the last row are the HDR cover images (tone mapped version of HDR cover images of ours, thus, also LDR cover images of the algorithms considered in comparison), stego images of our method, Chen et al.'s method [44], Van [45] and our method are almost as same as that of the cover images. On the contrary, the stego images of other four compared methods all have distortion in color although they are structurally similar to the cover images. ...
... It can be seen that in image pairs of our method, the overall trend of high frequency and low frequency information is basically unchanged, and the statistical characteristics of the generated images are still consistent with the original images. The histograms of the stego images of other methods are obviously different from that of the cover images, A deep learning-based steganography method for high dynamic range images Fig. 6 Results produced by HDR-VDP-2 using cover and stego images of different methods especially in the first row in Fig. 7. Besides, the histograms of the recovered secret images of Chen et al.'s method [44] and Van et al.'s method [45] are distinctly different from that of the original secret image. ...
High dynamic range (HDR) images have recently drawn much attention in multimedia community. In this paper, we proposed an HDR image steganography method based on deep learning, which is for HDR images with OpenEXR format. To the best of our knowledge, this is the first steganography method that applies deep learning to HDR image steganography, and the first steganography method that hides images in HDR images. The LDR secret image is hidden in the mantissa of the HDR cover image of the same size through a hidden network, and recovered through an extraction network in the receiver. Experimental results show that the proposed algorithm has advantages in security, robustness and capacity compared with other “hiding image in image” algorithms.
... This process is carried out carefully after proper training and optimization of the model parameters which vary from model to another. Theretofore, they are considered more compared to traditional techniques [6]. For examples DL techniques are utilized to perform end to end steganography. ...
... In steganographic applications, the use of CNNs is proposed using the encoder-decoder structure, where the former introduces the secret image and the latter extracts it. Several existing proposals start from this structure, with variations in terms of neural network architectures [19,20]. Other models apply CNNs to identify edges and borders within the images and use them to input messages [21]. ...
Multimedia content’s development and technological evolution have enhanced and even facilitated the application of steganography as a means to introduce hidden messages for cybercrime-related purposes. Artificial intelligence models have been widely implemented as a way to detect the presence of these messages in image content. However, the possibility of applying explainability techniques in order to provide visual representations of the signatures of different steganography algorithms has not been studied yet. This work presents a novel steganalysys methodology, STEG-XAI, not only for detecting steganography in images but also for explaining the machine learning model’s findings, and extracting the steganography algorithm’s signature. A convolutional neural network with EfficientNet architecture is implemented, along with the explainability algorithms LIME and Grad-CAM. The model is trained with a dataset of images modified by UERD, a steganography method designed for JPEG images, and achieves a weighted AUC of 0.944, displaying a high level of discrimination between original and tampered images. Furthermore, the explanation methods enable visualizing both the image modifications identified by the neural network, and a signature of the UERD algorithm.
... The secret and cover images are concatenated to give the input and hence 6 channels. A U-netbased Hiding (H-net) and revealing (R-net) network are used by authors in [14]. ...
... A new cost function to reduce the effect of noise in the generated container image called the variance loss is proposed [12]. An encoder-decoder architecture was proposed by Rahim et al. in [14]. This method differs from the others in the way the inputs are given. ...
... 14 [14] 2019 CNN based encoder-decoder architecture Payload capacity not high enough. ...
An image is the most popular media format amongst the current modern digital generation. Encoding binary data within an image is an easy way to hide the secret image. Broadly speaking, steganography is the practice of concealing a file, message, image, or video within another file, message, image, or video. Steganography helps us as the intended secret message does not attract attention to itself as an object of scrutiny. Plainly visible encrypted messages might be better protected but they arouse interest and may in themselves be incriminating in countries in which encryption is illegal. In other words, steganography is more discreet than cryptography when we want to send secret information while also being easier to extract. The usual implementations tend to significantly lose the image quality and are also easily detectable. However, this implementation makes efforts to overcome the existing problems of image steganography with the help of a deep neural network which results in the generation of a final image that is almost identical to the original image and isn’t detectable easily.
... Steganography method Improved Needs to be improved 2,3,[5][6][7][8] Traditional LSB based Easy implementation Security, payload capacity, visual quality of stego image and recovered image [9][10][11][12][13] Transform domain based Better security and payload capacity than traditional LSB Visual quality of stego and reconstructed images [14][15][16][17][18] Machine learning based Better visual quality of stego and reconstructed images High complexity, payload capacity can be improved 19,20 Support vector machine based Better security Not suitable for large dataset [23][24][25][26][27][28][29] CNN based High payload capacity, reconstruction quality Computational cost, security from deep learning based steganalysis 30 www.nature.com/scientificreports/ proposed work and its performance supremacy over the related works. ...
Increasing data infringement while transmission and storage have become an apprehension for the data owners. Even the digital images transmitted over the network or stored at servers are prone to unauthorized access. However, several image steganography techniques were proposed in the literature for hiding a secret image by embedding it into cover media. But the low embedding capacity and poor reconstruction quality of images are significant limitations of these techniques. To overcome these limitations, deep learning-based image steganography techniques are proposed in the literature. Convolutional neural network (CNN) based U-Net encoder has gained significant research attention in the literature. However, its performance efficacy as compared to other CNN based encoders like V-Net and U-Net++ is not implemented for image steganography. In this paper, V-Net and U-Net++ encoders are implemented for image steganography. A comparative performance assessment of U-Net, V-Net, and U-Net++ architectures are carried out. These architectures are employed to hide the secret image into the cover image. Further, a unique, robust, and standard decoder for all architectures is designed to extract the secret image from the cover image. Based on the experimental results, it is identified that U-Net architecture outperforms the other two architectures as it reports high embedding capacity and provides better quality stego and reconstructed secret images.
... Since two networks are to be trained, training these networks is a complex task. The training is reduced to by half with same accuracy in [53] as compared to [3]. In [53], H-net is used for embedding and R-net for reconstructing the secret image. ...
... The training is reduced to by half with same accuracy in [53] as compared to [3]. In [53], H-net is used for embedding and R-net for reconstructing the secret image. ...
Today, a lot of information is being shared electronically in a way or another. Despite the advancements in technology used for data transfer, the reliable transmission of sensitive data is still a major challenge that need to be addressed. In this paper, we propose a steganographic technique to generate a stego image using the Neural Style Transfer (NST) algorithm that maintain the perceptual quality of the stego image with maximum capacity payload. Along with this, we propose to recover the secret content from generated stego image with minimal distortion. So, to recover the secret image from the generated stego image, destylization is performed using conditional Generative Adversarial Networks (cGANs). The proposed destyling GAN is forced to learn the embedded secret information using a loss function that learns the same representation as in the embedding NST algorithm. This whole framework of embedding and extraction of secret image is evaluated for Imagenet dataset and later tested on the PASCAL VOC12 dataset. The algorithm outper-forms in terms of Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Visual Information Fidelity (VIF) with 44.175 dB, 0.9958, and 0.954 respectively for the Imagenet dataset. Also, the proposed algorithm is more robust against StegExpose with 0.529, area under the curve (AUC).
... It is based on a deep neural network, and it can also hide the same size image, which only modifies 0.76% cover image [26]. Van et al. proposed a new training scheme, which modifies the error back propagation to speed up the training of the network [24]. The above results use color images as secret information. ...
Color image steganography based on deep learning is the art of hiding information in the color image. Among them, image hiding steganography(hiding image with image) has attracted much attention in recent years because of its great steganographic capacity. However, images generated by image hiding steganography may show some obvious color distortion or artificial texture traces. We propose a color image steganographic model based on frequency sub-band selection to solve the above problems. Firstly, we discuss the relationship between the characteristics of different color spaces/frequency sub-bands and the generated image quality. Then, we select the B channel of the RGB image as the embedding channel and the high-frequency sub-band as the embedding domain. DWT(discrete wavelet transformation) transforms B channel information and secret gray image into frequency domain information, and then the secret image is embedded and extracted in the frequency domain. Comprehensive experiments demonstrate that images generated by our model have better image quality, and the imperceptibility is significantly increased.
... The secret and cover images are concatenated to give the input and hence 6 channels. A U-net based Hiding (H-net) and revealing (R-net) network are used by Van et al. in [28]. Batch normalization and ReLU activation are used. ...
... Architecture Dataset Advantages Disadvantages [31] Encoder-decoder ImageNet -Image is secret message -However image size is 64 × 64 which is very small [27] U-Net ImageNet -Image is secret message -Basic and minimum architecture is used -However image size is 64 × 64 which is very small -Input images are just concatenated [28] CNN ImageNet and Holiday -Image is secret message. Basic and minimum architecture is used -New error back propagation function is introduced to speed up training -However image size is 64 × 64 which is very small -Input images are just concatenated [29] Encoder-decoder with VGG base COCO and wikiart.org ...
... [16], [17], [47] and [10] have used MATLAB. [31], [54], [27] and [28] have used pytorch libraries from python. Another popular library used is the tensorflow [39], [53], [50] and [38]. ...
Image Steganography is the process of hiding information which can be text, image or video inside a cover image in a way that is not visible to the human eyes. Deep learning technology, which has emerged as a powerful tool in various applications including image steganography to protect and secure the transmitted data, has received increased attention recently. The main goal of this paper is to explore and discuss various deep learning methods available in image steganography field. Deep learning techniques used for image steganography can broadly be divided into three categories - traditional methods, CNN-based and GAN-based methods. Along with the methodology, an elaborate summary on the datasets used, experimental set-ups considered and the evaluation metrics commonly used are described in this paper. A table summarizing all the details are also provided for easy reference. This paper aims to help the fellow researchers by compiling the current trends, challenges and some future direction in this field.
