search

BioAITeam / GBRAS_SW

GBRAS_SW a spatial image steganalysis software.
1 stars 4 forks source link

Experimental results #1

Closed fennudehaogua closed 3 years ago

fennudehaogua commented 3 years ago

I feel a little confused about the experimental results in your paper, because there seems to be a big gap compared with the original paper. For example, in the original YeNet, SRNet paper, the data set is Boss+BOW2, and the detection error of WOW at 0.2bpp are 83.09% and 83.24%. The experimental results in your paper are 73.6% and 79.4%. Can you explain this problem and share your comparative experiment code (including YeNet, SRNet, ZhuNet, YEDROUDJ-NET)?Thank you very much！

BioAITeam commented 3 years ago

Hello I hope you are well When making the reproductions, the results may vary due to the great sensitivity of deep learning in this problem; for example, if you look at the new Zhu-Net article (link), you may see results that are not necessarily the same as the other articles. For example, in Ye-Net, SCA is not used...

We proposed improvements to the CNN architectures in the paper (link). Here you will find a repository. In which are the improvements to specific CNN's for steganalysis, and the application of the same strategy to other image classification networks such as VGG16.

I hope this will help.