Abstract:

Through analyzing the potential security problems of directly extending steganography based on the grayscale spatial image to the color spatial image, novel steganalysis features for detecting the steganographic methods with color spatial images as cover images were proposed. First of all, the Markov transition probability matr of the subtractive pixel adjacency matrices of three color channels were computed, and then the intra-color-channel features were extracted and merged. Afterwards, the transition probability mat ices of the subtractive pixel adjacency matrices among the color channels were calculated, where four horizontal and vertical directions, and four diagonal and anti-diagonal directions are separately merged. And the two parts composed the inter-color-channel features. Finally, the final feature set of the pro-posed method consists of the intra-color-channel features and the inter-color-channel features, and the ensemble classifier was used for the training and testing of the proposed feature. The experimental results show that the proposed method can effectively detect the steganographic schemes based on the color spatial image. Moreover, the proposed in-ter-color-channel features can efficiently capture the effect of the content-adaptive steganography on the correlation among the color channels of color images.

Key words: steganalysis, color spatial image, content-adaptive steganography, subtractive pixel adjacency matrix, stega-nalysis feature