Impact of feature selection in classification for hidden channel detection on the example of audio data hiding

ABSTRACT

The classification accuracy achieved in applied classification problems depends strongly on the choice of classifiers and, if a model based approach is chosen, on the quality of the model. In this paper, for a selected classification problem from the area of determination of the existence of hidden channels in audio data, the relevance of single features for model generation in a support vector machine based classification procedure is determined.

Here we consider nine audio data hiding algorithms as well as an existing audio steganalysis approach. The goal is to sharpen the model used for classification by algorithm specific generation of the feature set used and thereby reducing its dimensionality while keeping the same degree of classification accuracy for hidden channel detection on audio data. We show that for a multi-genre audio test set the impact of feature space reduction is less severe than for a set containing only speech. The fractions of the feature space considered significant in the performed multi-genre and speech evaluations (best results for the percentage of the available feature space considered significant in the tests performed here: 37.4% and 54.5% respectively) are determined for different thresholds of considering a feature significant in single feature classification. A first evaluation on embedding domain distinction is performed, distinguishing between time- and frequency/wavelet-domain.

The results for application specific steganalysis achieved here are compared to the results achieved in current image steganalysis schemes.

REFERENCES

• 1: Pevny, T.; and Fridrich, J.: Merging Markov and DCT features for multi-class JPEG steganalysis. Electronic Imaging Conf. on Security, Steganography, and Watermarking of Multimedia Contents, 2007.
• 2: Avcibas, I.; Memon, N. and Sankur, B.: Steganalysis using Image Quality Metrics; Electronic Imaging Conf. on Security and Watermarking of Multimedia Contents, 2001.
• 3: David J. Hand , Padhraic Smyth , Heikki Mannila, Principles of data mining, MIT Press, Cambridge, MA, 2001
• 4: S. Kung , M. Mak , S. Lin, Biometric authentication: a machine learning approach, Prentice Hall Press, Upper Saddle River, NJ, 2004
• 5: Miche, Y.; Roue, B.; Lendasse, A. and Bas, P.: A feature selection methodology for steganalysis. In Proceedings of the International Workshop on Multimedia Content Representation, Classification and Security, Springer, 2006.
• 6: Avcibas, I.: Audio steganalysis with contentindependent distortion measures. IEEE Signal Processing Letters, Vol. 13, No. 2, 2006.
• 7: Solanki, K.; Sarkar, A; and Manjunath, B. S.: YASS: Yet Another Steganographic Scheme That Resists Blind Steganalysis. Proceedings of Information Hiding 2007.
• 8: Ozer H.; Avcibas I.; Sankur B.; and Memon N.: Steganalysis of audio based on audio quality metrics. SPIE Electronic Imaging Conf. on Security and Watermarking of Multimedia Contents, 2003.
• 9: Kraetzer, C. and Dittmann, J.: Pros and Cons of Melcepstrum based Audio Steganalysis using SVM Classification. Proceedings of Information Hiding 2007.
• 10: Kludas, J.; Bruno, E.; and Marchand-Maillet, S.: Can Feature Information Interaction help for Information Fusion in Multimedia Problems? First International Workshop on Metadata Mining for Image Understanding, 2008.
• 11: Chang, C.-C. and Lin, C.-J.: LIBSVM: a Library for Support Vector Machines. 2001
• 12: Johnson, M. .; Lyu, S.; and Farid, H.: Steganalysis of recorded speech. Electronic Imaging Conf. on Security and Watermarking of Multimedia Contents, 2005.
• 13: Ru, X.-M.; Zhang, H.-J.; and Huang, X.: Steganalysis of audio: Attacking the steghide. Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, 2005.
• 14: Kraetzer, C.; and Dittmann, J.: Cover Signal Specific Steganalysis: the Impact of Training on the Example of two Selected Audio Steganalysis Approaches. To appear in Proceedings of the Electronic Imaging Conf. on Security, Forensics, Steganography, and Watermarking of Multimedia Contents, 2008.

INDEX TERMS

Primary Classification:
  I. Computing Methodologies
  I.5 PATTERN RECOGNITION
      I.5.2 Design Methodology

General Terms:
Measurement, Performance, Reliability, Verification

Keywords:
audio steganalysis, feature selection