A Blind Detection Method for Additive Noise Steganography in JPEG Decompressed Images
Li Xing, Tao Zhang, Kaida Li, Xijian Ping · 2011
Based on the variance analysis of the noise residuals in the DCT domain, a novel blind steganalyzer is proposed for additive noise steganography in JPEG decompressed images. After investigating the influence of the data embedding on the variance and statistical distribution of DCT ac coefficients, we extract a one-dimensional feature which is the area ratio of the coefficient normalized histogram in different range. Extensive experiments on LSB matching, ±K embedding and Stochastic Modulation Steganography, as well as the comparisons with prior art demonstrate that the proposed scheme can effectively detect the presence of the secret message even for a very low embedding rate. And it significantly outperforms the existing methods. Moreover, our method is practical and real-time.