Effective Steganalysis Based on Statistical Moments of Differential Characteristic Function

Based on statistical moments of differential characteristic function, an effective steganalytic technique is proposed in this paper. It calculates the first, second and third order partial differentiations, the first and second order total differentiations at pixel-locations in the test image. Eighteen statistics were computed from these five objects and intensity in image utilizing histogram and co-occurrence matrix. For each statistic, the characteristic function is calculated. The first and second statistical moments of the characteristic functions from all the statistics are selected to form a 60-dimensional feature vector for steganalysis. Support vector machine was utilized as classification algorithm. Compared with known schemes, the presented method demonstrates higher detecting rates with lower false positives for Cox and Piva spread spectrum steganographies