AI helps you reading Science
AI generates interpretation videos
AI extracts and analyses the key points of the paper to generate videos automatically
AI parses the academic lineage of this thesis
AI extracts a summary of this paper
Robust Steganalysis Based On Training Set Construction And Ensemble Classifiers Weighting
2015 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP), pp.1498-1502, (2015)
- Modern image steganalysis is usually considered as a supervised classification problem, in which images are represented by feature vectors and a classifier as steganalyzer is trained on cover and stego examples.
- When the testing images are from a source very different from the training set, the performance often degrades significantly
- This has been called the cover source mismatch problem in recent literature [1,2], and has been recognized as a serious issue that may complicate deployment of steganalyzers in real world.
- The time and space complexity may be too high and the usefulness of some data is questionable
- Modern image steganalysis is usually considered as a supervised classification problem, in which images are represented by feature vectors and a classifier as steganalyzer is trained on cover and stego examples
- The problem is that it is difficult to get a reliable division on a large data set with high diversity of samples. Another approach focuses on centering or normalizing samples in the feature space [1, 6], but the improvement is limited because the variation of samples is irregularly affected by so many factors
- To construct a large representative training set, we propose using two criteria to select samples from a large amount of images on the Internet: 1) the distance from a sample to previous classifiers learned on the initial training set and, 2) the diversity of the selected samples
- We have proposed to use large representative training set and weighted ensemble classifiers to improve the classification robustness to high intra-class variation
- Experimental results show that our method can improve the performance and robustness of steganalysis under high intra-class variation
- The proposed weighted ensemble classifiers, which are adaptive to the location of testing sample in the feature space, have been proved to be effective in the experiments
- The authors' overall goal is to construct a training set with samples of high variation and low redundancy.
- In the construction of training set, the authors try to select diverse and representative samples based on the idea of active learning and instance selection .
- To make the final constructed training set more compact, the authors employ a diversity measure to further reduce the redundancy of the selected samples
- Experimental results show that the method can improve the performance and robustness of steganalysis under high intra-class variation.
- By enlarging BOSSbase using image processing operations, the detection errors decrease quickly by more than 10%
- The steganalyzer needs to be robust to various images from different sources.
- The authors have proposed to use large representative training set and weighted ensemble classifiers to improve the classification robustness to high intra-class variation.
- The proposed weighted ensemble classifiers, which are adaptive to the location of testing sample in the feature space, have been proved to be effective in the experiments.
- The authors will do more experiments on different steganographic algorithms using massive image data.
- The impact of the original database, the setting of the parameters λ, T, N , as well as the treatment for “unsure” samples will be further studied
- Table1: Detection errors of classifiers on nsF5 0.2 bpnc. Feature space: LIU (216)
- Table2: Detection errors of classifiers on nsF5 0.2 bpnc. Feature space: CC-PEV (548)
- This work is funded by the National Basic Research Program of China (Grant No 2012CB316300), the National Nature Science Foundation of China (Grant No.61303262), and the National Key Technology R&D Program (Grant No.2012BAH04F02)
- Andrew D. Ker and T. Pevny, “A mishmash of methods for mitigating the model mismatch mess,” in Proceedings SPIE, Media Watermarking, Security, and Forensics, 2014, vol. 9028, p. 90280I.
- J. Kodovsky, V. Sedighi, and J. Fridrich, “Study of cover source mismatch in steganalysis and ways to mitigate its impact,” in Proceedings SPIE, Media Watermarking, Security, and Forensics, 2014, vol. 9028, p. 90280J.
- G. Cancelli, G. Doerr, M. Barni, and Ingemar J. Cox, “A comparative study of steganalyzers,” in Multimedia Signal Processing, 2008 IEEE 10th Workshop on, 2008, pp. 791–796.
- T. Pevny, P. Bas, and J. Fridrich, “Steganalysis by subtractive pixel adjacency matrix,” IEEE Transactions on Information Forensics and Security, vol. 5, pp. 215–224, 2010.
- M. Barni, G. Cancelli, and A. Esposito, “Forensics aided steganalysis of heterogeneous images,” in IEEE International Conference on Acoustics Speech and Signal Processing, 2010, pp. 1690–1693.
- X. Li, X. Kong, B. Wang, Y. Guo, and X. You, “Generalized transfer component analysis for mismatched JPEG steganalysis,” in IEEE International Conference on Image Processing, 2013, pp. 4432–4436.
- Ivans Lubenko and Andrew D. Ker, “Steganalysis with mismatched covers: Do simple classifiers help,” in Proceedings of the on Multimedia and Security. 2012, pp. 11–18, ACM.
- Ivans Lubenko and Andrew D. Ker, “Going from small to large data in steganalysis,” in Proceedings SPIE, Media Watermarking, Security, and Forensics, 2012, vol. 8303, p. 83030M.
- C. Chen and Y.Q. Shi, “JPEG image steganalysis utilizing both intrablock and interblock correlations,” in Proceedings of IEEE International Symposium on Circuits and Systems, 2008, pp. 3029–3032.
- J. Fridrich and J. Kodovsky, “Rich models for steganalysis of digital images,” Information Forensics and Security, IEEE Transactions on, vol. 7, pp. 868–882, 2012.
- J. Kodovskyand J. Fridrich, “Steganalysis of JPEG images using rich models,” in Proceedings SPIE, Electronic Imaging, Media Watermarking, Security, and Forensics XIV, 2012, vol. 8303.
- Mattis Paulin, Jerome Revaud, Zaid Harchaoui, Florent Perronnin, Cordelia Schmid, et al., “Transformation pursuit for image classification,” in IEEE Conference on Computer Vision & Pattern Recognition, 2014.
- H. Liu and H. Motoda, “On issues of instance selection,” Data Mining and Knowledge Discovery, vol. 6, no. 2, pp. 115–130, 2002.
- J. Kodovskyand J. Fridrich, “Effect of image downsampling on steganographic security,” Information Forensics and Security, IEEE Transactions on, vol. 9, no. 5, pp. 752–762, 2014.
- J. Kodovskyand J. Fridrich, “Steganalysis in resized images,” in IEEE International Conference on Acoustics, Speech and Signal Processing, 2013, pp. 2857–2861.
- “BOSS1.01,” http://exile.felk.cvut.cz/boss/BOSSFinal/index.php.
- J. Kodovsky, J. Fridrich, and V. Holub, “Ensemble classifiers for steganalysis of digital media,” Information Forensics and Security, IEEE Transactions on, vol. 7, pp. 432–444, 2012.
- Klaus Brinker, “Incorporating diversity in active learning with support vector machines,” in Proceedings of International Conference on Machine Learning, 2003.
- J. Fridrich, T. Pevny, and J. Kodovsky, “Statistically undetectable JPEG steganography: Dead ends, challenges, and opportunities,” in Proceedings of the 9th workshop on Multimedia and security. 2007, pp. 3–14, ACM.
- Q. Liu, “Steganalysis of DCT-embedding based adaptive steganography and yass,” in Proceedings of the Thirteenth ACM Multimedia Workshop on Multimedia and Security, New York, NY, USA, 2011, pp. 77–86, ACM.
- T. Pevnyand J. Fridrich, “Merging markov and DCT features for multi-class JPEG steganalysis,” 2007, vol. 6505.
- J. Kodovskyand J. Fridrich, “Calibration revisited,” in Proceedings of the 11th ACM Workshop on Multimedia and Security. 2009, pp. 63–74, ACM.
- “Source codes,” http://dde.binghamton.edu/download/.