Steganalysis in Multiple Sources of Cover Images

45 dimensional feature vectors were extracted from DCT and spatial domains of images to defeat steganographic schemes in this paper. Compared with commonly used steganalytic algorithms, the presented method offers much better performances in double JPEG compressed covers and their stegos. However, when our method was applied to detect whether multiple sources of images containing hidden data or not, it showed very poor performances. An improved training technique was proposed and a robuster steganalytic algorithm was designed.