Feature-Based Steganalysis for JPEG Images

The goal of blind steganalysis is to detect the presence of hidden data and to eventually extract them from the stego images generated by various data hiding schemes. In this paper, we construct a new blind classifier capable of detecting several steganographies for JPEG images. Thirteen statistics are collected in the DCT domain and spatial domain. By using the characteristic function and the center of mass (COM) for each statistic, we calculate an 82-dimensional feature vector for each image. Support Vector Function (SVM) is utilized to construct the blind classifier. Experimental results show that the proposed steganalytic method provides significant performance on various types of steganographies, such as Model-based steganography MB1[17]&MB2[18], non-blind spread spectrum data hiding method Cox[16], and five popular data hiding schemes-F5[11], JSteg[12], Jphide&Seek[13], Outguess[14] and Steghide[15].