Investigation of steganalysis performance of selected steganography methods with deep learning models

Basit Öğe Kaydı
dc.contributor.authorBulus, Ercan
dc.date.accessioned2024-10-29T17:58:57Z
dc.date.available2024-10-29T17:58:57Z
dc.date.issued2024
dc.departmentTekirdağ Namık Kemal Üniversitesi
dc.description.abstractPurpose:It is to research whether steganalysis can be done with deep learning methods. Theory and Methods: 10,000 BOWS 2 and 4,000 BOSSBase 1.01 pairs were used for training, 1,000 BOSSBase 1.01 pairs forvalidation, and 5,000 BOSSBase 1.01 pairs for testing. HILL, MiPOD, S-UNIWARD and WOW steganography methods were used to information hiding. Steganalysis was performed with VGG16, Xu-Net, Ye-Net and Yeroudj-Net Deep learning models. Results: At 0.2 bpp load there is not much difference between the results and not quite as expected. The best result in wow at 0.4 payload was obtained with the VGG16 (86.4%) method. However, VGG16 test times are twice as long compared to others. In this case, the Yedroudj method seems more suitable in terms of accuracyrate and processing time. Conclusion: When the test accuracy results for 0.2bpp load are examined in the study; It is seen that the lowest result is 50% (0.500) in the YEDROUDJ model with the MIPOD method, and the most efficient result is 69% (0.69) in the XU-net model for all methods. Since acceptable results could not be achieved with the VGG16 model under 0.2bpp load, it was not evaluated. In this case, the Xu-net model can be preferred to investigate whetherinformation is hidden in images with a low amount of information such as 0.2bpp. Again, when the test accuracy results for 0.4bpp load are examined in the study; It is seen that the lowest result is obtained withthe Xu-net model in the MIPOD method with 72% (0.72) and again with the Ye-Net model in the MIPOD method with 72% (0.72). On the other hand, the highest result was obtained with the VGG16 model in the WOW method with 86% (0.864) and with the VGG16 model with 84% (0.847). However, when the processing times in Table 11 are examined, it is clear that the processing time of the examinations made with the VGG16 model is almost twice as long as the others, and this is a very undesirable situation. In this case, the Yedroudj method is seen as a more useful method since it has the second best accuracy rate between 75%-82% with 0.4bpp load and a lower processing time (half) than VGG16.
dc.identifier.doi10.17341/gazimmfd.1324765
dc.identifier.issn1300-1884
dc.identifier.issn1304-4915
dc.identifier.issue1en_US
dc.identifier.scopus2-s2.0-85201687074
dc.identifier.scopusqualityQ2
dc.identifier.urihttps://doi.org/10.17341/gazimmfd.1324765
dc.identifier.urihttps://hdl.handle.net/20.500.11776/14571
dc.identifier.volume40
dc.identifier.wosWOS:001309399500009
dc.identifier.wosqualityN/A
dc.indekslendigikaynakWeb of Science
dc.indekslendigikaynakScopus
dc.language.isotr
dc.publisherGazi Univ, Fac Engineering Architecture
dc.relation.ispartofJournal of The Faculty of Engineering and Architecture of Gazi University
dc.relation.publicationcategoryMakale - Uluslararası Hakemli Dergi - Kurum Öğretim Elemanıen_US
dc.rightsinfo:eu-repo/semantics/openAccess
dc.subjectImage Steganography
dc.subjectImage Steganalysis
dc.subjectData Hiding
dc.subjectDeep Learning Networks
dc.subjectCryptography
dc.titleInvestigation of steganalysis performance of selected steganography methods with deep learning models
dc.typeArticle

Dosyalar

Koleksiyon

WoS İndeksli Yayınlar Koleksiyonu
Scopus İndeksli Yayınlar Koleksiyonu