Heterogeneous Image Steganalysis Based on Forensics Assistance

Heterogeneous Image Steganalysis Based on Forensics Assistance

연구 분야: Strategies

학회: 2024 6th International Conference on Communications, Information System and Computer Engineering (CISCE)

The purpose of image steganalysis is to detect secret information embedded in digital images. Currently, deep learning-based image steganalysis methods have achieved good detection performance. However, existing deep learning-based steganalysis methods rarely consider image selection. And some networks only use fixed filter kernels, rarely exploiting the learning ability of the network. This leads to the mismatch problem of existing steganalysis models and poses great challenges to steganalysis. To solve the mismatch problem of hybrid heterogeneous image steganalysis, we propose a forensic-assisted image steganalysis scheme. First, a classifier is used to classify the image source and filter different types of image sets, and then different versions of the steganalysis analyser are trained for different types of image sets to process the corresponding classes of images. The Steganalyser is designed to perform its feature learning efficiently. In the experiments, we applied the most advanced steganalysis model and several datasets to validate the scheme and achieve better results. The feasibility of the scheme is verified.