Toward Ternary Classification in CNN-Based Image Steganalysis

초록

This paper presents a study for ternary (cover,WOW-stego, and UNIWARD-stego) classification in CNN-basedimage steganalysis. A conventional steganalytic CNN is extendedto identify different steganographic methods applied to input images. In addition, a new preprocessing high-pass filter isproposed to extract steganalytic features from input images,which uses the SRM filters with additional operations.Experimental results showed that the proposed method could notonly identify whether a steganographic method has been applied to input image, but also what stegano graphic method is with high accuracy.