Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond
Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond
IHW '01 Proceedings of the 4th International Workshop on Information Hiding
A Classification Framework for Anomaly Detection
The Journal of Machine Learning Research
Estimation of High-Density Regions Using One-Class Neighbor Machines
IEEE Transactions on Pattern Analysis and Machine Intelligence
Estimating the Support of a High-Dimensional Distribution
Neural Computation
Defending against statistical steganalysis
SSYM'01 Proceedings of the 10th conference on USENIX Security Symposium - Volume 10
Statistically undetectable jpeg steganography: dead ends challenges, and opportunities
Proceedings of the 9th workshop on Multimedia & security
A Markov process based approach to effective attacking JPEG steganography
IH'06 Proceedings of the 8th international conference on Information hiding
Modified matrix encoding technique for minimal distortion steganography
IH'06 Proceedings of the 8th international conference on Information hiding
YASS: yet another steganographic scheme that resists blind steganalysis
IH'07 Proceedings of the 9th international conference on Information hiding
A graph–theoretic approach to steganography
CMS'05 Proceedings of the 9th IFIP TC-6 TC-11 international conference on Communications and Multimedia Security
Towards multi-class blind steganalyzer for JPEG images
IWDW'05 Proceedings of the 4th international conference on Digital Watermarking
IH'04 Proceedings of the 6th international conference on Information Hiding
Steganalysis using higher-order image statistics
IEEE Transactions on Information Forensics and Security
A method for automatic identification of signatures of steganography software
IEEE Transactions on Information Forensics and Security
BSS: Boosted steganography scheme with cover image preprocessing
Expert Systems with Applications: An International Journal
Moving steganography and steganalysis from the laboratory into the real world
Proceedings of the first ACM workshop on Information hiding and multimedia security
| Hi-index | 0.00 |
It is generally believed that a blind steganalyzer trained on sufficiently many diverse steganographic algorithms will become universal in the sense that it will generalize to previously unseen (novel) stego methods. While this is a partially correct statement if the embedding mechanism of the novel method resembles some of the methods on which the classifier was trained, we demonstrate that if the classifier is presented with stego images produced by a completely different embedding mechanism, it may fail to detect the images as stego even for an otherwise fairly easily detectable method. Motivated by this observation, we explore two approaches for construction of universal steganalyzers - one-class and one-against-all classifiers. Their advantages and disadvantages are discussed and performance compared on a wide variety of steganographic algorithms. One-against-all classifiers have generally better performance than approaches based on characterizing just the class of covers but they may fail catastrophically on previously unseen stego algorithms. One-class methods are less likely to fail to detect unknown stego algorithms but have lower overall detection accuracy on known stego methods. The suitability of each approach thus depends on the application.