Efficient steganographer detection over social networks with sampling reconstruction

Abstract

This work proposes an improvement solution in identifying malicious user (or steganographer) who try to deliver hidden information in a batch of natural images. In this solution, a sampling construction strategy is proposed firstly. We design a probability calculation model by analying the principle of adaptive steganography, and then select DCT blocks with higher embedding probability to reconstruct a sample image, which is considered as the proof of extracting steganalysis features. Furthermore, inspired by the classical PEV-193 feature space, we reform a reduced PEV feature set including histogram features and intra-block co-occurrence features, which can capture more steganographic changes and match the sampling construction strategy well. Comprehensive experimental results show that comparing with the state-of-the-arts, the proposed scheme has a significant improvement in identifying potential steganographers in large-scale social media networks, and therefore is believed to be able to resist adaptive steganography with small payload.