JPEGCNN: A Transform Domain Steganalysis Model Based on Convolutional Neural Network

Abstract

Convolutional Neural Network (CNN) has gained an overwhelming advantage in many fields of pattern recognition. Both excellent data learning ability and automatic feature extraction ability of CNN are urgently needed in image steganalysis. However, the application of CNN in image steganalysis is still in its infancy, especially in the field of JPEG steganalysis. In this paper, a steganalysis model based on CNN in gray image transform domain is proposed, which is called JPEGCNN. At the same time, on the basis of JPEGCNN, JPEGCNN is extended to the transform domain of color image by researching and designing different methods of feature extraction. RGBMERGE-JPEGCNN and RGBADD-JPEGCNN are proposed respectively, which make up for the lack of research on steganalysis model based on convolution neural network in the transform domain of color image. Experiments show that JPEGCNN, RGBMERGE-JPEGCNN and RGBADD-JPEGCNN proposed in this paper have good detection ability for steganography algorithm in transform domain.