Universal Steganalysis Using Color Correlation and Feature Fusion

Abstract

A new universal steganalysis algorithm using YUV color space conversion and feature combination of DCT (discrete cosine transform) and DWT (discrete wavelet transform) domains is proposed. It follows the philosophy introduced in in which the features are calculated from the luminance and chrominance components of the images. Instead of using single wavelet high-order statistics for luminance component of the image, it extracts features both in DCT and DWT domains, and substitute the wavelet high-order statistics with the moments of wavelet characteristic function which is more sensitive to steganography. Then a nonlinear SVM classifier is trained on a database of images to construct a universal steganalyzer. Applying the scheme to detect the steg images produced by Jsteg, F5 and Outguess, the experimental results have shown the proposed algorithm get an average detection accuracy of 84.1%, indicate that this method offers improved performance.