GBRAS-Net: A Convolutional Neural Network Architecture for Spatial Image Steganalysis

Tabares-Soto, Reinel; Arteaga-Arteaga, Harold Brayan; Bravo-Ortíz, Mario Alejandro; Mora-Rubio, Alejandro; Arias-Garzón, Daniel; Alzate-Grisales, Jesús Alejandro; Buenaventura, Burbano-Jacome Alejandro; Orozco-Arias, Simón; Isaza, Gustavo; Raul, Ramos-Pollan

doi:10.1109/access.2021.3052494

Cited by 56 publications

(41 citation statements)

References 28 publications

(43 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After the corresponding steganographic images are generated, the BOSSBase 1.01 database contains 10,000 pairs of images (cover and stego) divided into 4,000 pairs for train, 1,000 pairs for validation, and 5,000 for the test. This partition of the BOSSBase 1.01 database was based on Xu, Wu & Shi (2016) , Ye, Ni & Yi (2017) , Yedroudj, Comby & Chaumont (2018) , Boroumand, Chen & Fridrich (2019) , Zhang et al, 2019 , Reinel et al (2021) .…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Sensitivity of deep learning applied to spatial image steganalysis

Tabares-Soto

Arteaga-Arteaga

Mora-Rubio

et al. 2021

PeerJ Computer Science

Self Cite

View full text Add to dashboard Cite

In recent years, the traditional approach to spatial image steganalysis has shifted to deep learning (DL) techniques, which have improved the detection accuracy while combining feature extraction and classification in a single model, usually a convolutional neural network (CNN). The main contribution from researchers in this area is new architectures that further improve detection accuracy. Nevertheless, the preprocessing and partition of the database influence the overall performance of the CNN. This paper presents the results achieved by novel steganalysis networks (Xu-Net, Ye-Net, Yedroudj-Net, SR-Net, Zhu-Net, and GBRAS-Net) using different combinations of image and filter normalization ranges, various database splits, different activation functions for the preprocessing stage, as well as an analysis on the activation maps and how to report accuracy. These results demonstrate how sensible steganalysis systems are to changes in any stage of the process, and how important it is for researchers in this field to register and report their work thoroughly. We also propose a set of recommendations for the design of experiments in steganalysis with DL.

show abstract

Section: Methodsmentioning

confidence: 99%

“…Currently, GBRAS-Net architecture, presented by Reinel et al (2021) , achieves the highest detection percentages of steganographic images in the spatial domain. In the preprocessing stage, this network keeps the 30 SRM filters and uses a modified TanH activation function.…”

Section: Introductionmentioning

confidence: 99%

Sensitivity of deep learning applied to spatial image steganalysis

Tabares-Soto

Arteaga-Arteaga

Mora-Rubio

et al. 2021

PeerJ Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…The most recent contribution was published by Reinel et al (2021). This network maintains, for the preprocessing stage, the 30 SRM filters and has a 3 × TanH activation function.…”

Section: Introductionmentioning

confidence: 99%

Strategy to improve the accuracy of convolutional neural network architectures applied to digital image steganalysis in the spatial domain

Tabares-Soto

Arteaga-Arteaga

Mora-Rubio

et al. 2021

PeerJ Computer Science

Self Cite

View full text Add to dashboard Cite

In recent years, Deep Learning techniques applied to steganalysis have surpassed the traditional two-stage approach by unifying feature extraction and classification in a single model, the Convolutional Neural Network (CNN). Several CNN architectures have been proposed to solve this task, improving steganographic images’ detection accuracy, but it is unclear which computational elements are relevant. Here we present a strategy to improve accuracy, convergence, and stability during training. The strategy involves a preprocessing stage with Spatial Rich Models filters, Spatial Dropout, Absolute Value layer, and Batch Normalization. Using the strategy improves the performance of three steganalysis CNNs and two image classification CNNs by enhancing the accuracy from 2% up to 10% while reducing the training time to less than 6 h and improving the networks’ stability.

show abstract

“…Modern methods for covert (steganographic) communication are aimed at message embedding into innocuous (cover) files, such as digital images (DI), by preserving low level of cover's features alteration [2,3]. Detection of formed stego images requires usage of either statistical models [4], or convolutional neural networks [5,6] images even under limited prior information about used embedding method is topical task. One of promising approaches for solving mentioned task is DI pre-processing (calibration) for emphasizing weak alterations caused by message hiding [7].…”

Section: Introductionmentioning

confidence: 99%

Effectiveness of stego images pre-noising with fractional noise for digital image steganalysis

Progonov¹

2021

AAIT

View full text Add to dashboard Cite

Counteraction to sensitive data leakage in cyber-physical systems is topical task today. Solving of the task is complicated to widely usage by attackers of novel steganographic methods for sensitive data embeddinginto innocuous (cover)files, such as digital images. Feature of these embedding methods isminimization of cover image’s parameters alterations during message hiding. This negatively affectsdetection accuracy of formed stego images bystate-of-the-art statistical stegdetectors. Therefore, advanced methods for detection and amplification of cover image’s parameters abnormal changes caused by data embedding are needed.The novel approach for solving of mentioned task is applying of image pre-processing (calibration) methods. These methods are aimed at estimation parameters either of cover, or stego images from current analysed image.The majority of known calibration methods are based on cover image content suppression by utilization of extensive set of high-pass filters. This makes possible close to state-of-the-art detection accuracy by the costof time consuming preselectionof appropriate filters. Therefore, this approach may be inappropriate in real cases, when fast re-train stegdetector for revealing of stego images formed by unknown embedding methodsis required.For overcoming this limitation, we proposed to calibrate an image by amplification of alterations caused by message hiding. This can be realized by data re-embedding into images or their pre-noising.The effectiveness of such approach was proved for wide range of modern embedding methods in the case of message re-embedding. The paper is aimed at performance analysis of image calibration by pre-noising, namely by using of non-stationary fraction noise.The performance analysis of proposed solution was performed for novel HUGO and MG adaptive embedding methods on standard VISION dataset. According to obtained results, we may conclude that applying of proposed solution allows achieving close to state-of-the-art detection accuracy for HUGO embedding method and low (less than 10%) cover image payload.Also, low computation complexity of proposed solution makes it an attractive alternative to novel cover rich models based stegdetectors. Nevertheless, solution’s performance concedes effectiveness of novel stegdetectorsfor medium(less than 20%) and high (more 25%) cover image payloadfor MG embedding method

show abstract

GBRAS-Net: A Convolutional Neural Network Architecture for Spatial Image Steganalysis

Cited by 56 publications

References 28 publications

Sensitivity of deep learning applied to spatial image steganalysis

Sensitivity of deep learning applied to spatial image steganalysis

Strategy to improve the accuracy of convolutional neural network architectures applied to digital image steganalysis in the spatial domain

Effectiveness of stego images pre-noising with fractional noise for digital image steganalysis

Contact Info

Product

Resources

About