An efficient JPEG steganalysis scheme based on Binary Coded Genetic Algorithm and cognitive ensemble classifier

In this paper, we propose an efficient steganalysis method by using Cartesian calibrated JPEG Rich Models (ccJRM) features set. Proposed steganalysis scheme contains two steps: 1) search a subset of features (among set of 22510 features) with the most promising performances, and 2) build an cognitive ensemble classifier for efficient steganalysis. In the first step we used Binary Coded Genetic Algorithm (BCGA) coupled with Extreme Learning machine to collect few subset of features with promising performances and corresponding ELM models. In the second step we used another BCGA for searching the best combination of few ELM models computed in the first step. Chosen combination of ELM models is used to build a cognitive ensemble classifier. Proposed steganalysis scheme shows an improvement compared to existing JPEG steganalysis schemes.