A CNN-Based HEVC Video Steganalysis Against DCT/DST-Based Steganography

Zhenzhen Zhang, Henan Shi, Xinghao Jiang, Zhaohong Li, Jindou Liu

Abstract: The development of video steganography has sparked ever-increasing concerns over video steganalysis. In this paper, a novel steganalysis approach against Discrete Cosine/Sine Transform (DCT/DST) based steganography for High Efficiency Video Coding (HEVC) video is proposed. The distortion of DCT/DST-based HEVC steganography and the impact on pixel value of HEVC videos is firstly analyzed. Based on the analysis, a convolutional neural network (CNN) is designed. The proposed CNN is mainly composed of three parts, i.e. residual convolution layer, feature extraction and binary classification. In the feature extraction part, a steganalysis residual block module and a squeeze-and-excitation (SE) block are designed to improve the network’s representation ability. In comparison to the existing steganalysis methods, experimental results show that the proposed network performs better to detect DCT/DST-based HEVC steganography.