Volume 2, Issue 4 (12-2010)                   IJICTR 2010, 2(4): 1-8 | Back to browse issues page

XML Print


Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:

Pezhmanpour M, Sheikhan M, Moin M S. Blind Color Image Steganalysis Based on Multi-Transform Combined Features Selected by a Hybrid of ANOVA and BPSO . IJICTR. 2010; 2 (4) :1-8
URL: http://ijict.itrc.ac.ir/article-1-245-en.html
1- EE Dept. Islamic Azad University­ South Tehran Branch Tehran, Iran
2- Multimedia Sys. Group IT Dept. Iran Telecom Research Center Tehran, Iran
Abstract:   (1382 Views)
Steganalysis techniques have been classified into two major categories: blind steganalysis which is independent of the steganography method and specific steganalysis which attempts to detect specific steganographic media. Feature extraction is an important functional block in the steganalysis systems. The features are commonly in spatial domain or extracted from transform domains such as discrete wavelet transform (DWT), discrete cosine transform (DCT) or contourlet transform (CT). In this paper, a blind color image steganalysis method based on a hybrid set of features (statistical moments, entropy, and co-occurrence matrix features) extracted from a combination of DWT, DCT, and CT is proposed. The hybrid of "analysis of variation (ANOVA)" as an open-loop feature selection method, and "binary particle swarm optimization (BPSO)" as a closed-loop one, is used in this work to improve the detection rate in tandem with significant reduction in the size of feature set. Jsteg, OutGuess, JPHS and model-based steganography methods are attacked in this work. By using the hybrid of "ANOVA+BPSO", the number of features is reduced to 13. Empirical results show that the most discriminative features in clean/stego image classification are statistical moments of co-occurrence matrix of contourlet transform. The most discriminative selected features are fed into a nonlinear support vector machine (SVM) classifier to distinguish the cover and stego images. Average detection accuracy of the proposed model is above 81 percent for the embedding-rate ranges of 5% to 25%.
Full-Text [PDF 1214 kb]   (550 Downloads)    
Type of Study: Research | Subject: Information Technology

Add your comments about this article : Your username or Email:
CAPTCHA code