Rich model for Steganalysis of color images

Abstract

In this paper, we propose an extension of the spatial rich model for steganalysis of color images. The additional features are formed by three-dimensional co-occurrences of residuals computed from all three color channels and their role is to capture dependencies across color channels. These CRMQ1 (color rich model) features are extremely powerful for detection of steganography in images that exhibit traces of color interpolation. Content-adaptive algorithms seem to be hurt much more because of their tendency to modify the same pixels in each channel. The efficiency of the proposed feature set is demonstrated on three different color versions of BOSSbase 1.01 and two steganographic algorithms - the non-adaptive LSB matching and WOW.