Fuzzy Based Clustering for Grayscale Image Steganalysis


Steganography is the science that involves communicating secret message in a multimedia carrier. On the other hand, steganalysis is the field dedicated to detect whether a given multimedia has hidden message in it. The detection of hidden messages is revealed as a classification problem. To this end, this paper has two contributions. Up to the best of our knowledge, this is the first time todefine grayscale image steganalysis, as a fuzzy c-means clustering (FCM) problem. The objective of the formulated fuzzy problem is to construct two fuzzy clusters: cover-image and stego-image clusters. The second contribution is to define a new detector, called calibrated Histogram Characteristic Function (HCF) with HaarWavelet(HCF^HW). The proposed detector is exploited, by the fuzzy clustering algorithm, as a feature set parameter to define the boundaries of the cover- and stego- images clusters. Performance evaluations of FCM with HCF^HW in terms of accuracy, detection rate, and false positive rate are investigated and compared with other work based on HCF Center of Mass or HCF-COM andcalibrated HCF-COM by down sampling. The comparison reveals out that the proposed FCM with (HCF^Hw)significantly outperforms other work.