Binary Coded Genetic Algorithm with Ensemble Classifier for feature selection in JPEG steganalysis

In this paper, we propose a Binary Coded Genetic Algorithm with Ensemble Classification feature selection procedure designed for steganalysis. Proposed feature selection method was used for searching the most appropriate subset of features from 22510 dimension feature space superior for JPEG steganalysis. Reduced set of features shows better classification accuracy for JPEG steganalysis compared to complete set of features. In our method we used an ensemble classifier to approximate the functional relationship between the reduced feature set and class label. Search for optimal subset of features requires to solve two optimization problems: define the optimal number of features and define the optimal subset itself. Proposed Binary Coded Genetic algorithm enables to solve two optimization problems together. Each feature is coded as a binary coefficient in a binary string, which represent one solution of the feature selection problem. Genetic operations executed for binary strings (parents) results new binary strings (child) with good chance to have higher classification accuracy for JPEG steganalysis. Experimental results clearly indicate the advantage of using the proposed reduced set of features for JPEG steganalysis.