Evaluation of Feature Selection Measures for Steganalysis

Abstract

Steganalysis has attracted researchers' attention overwhelmingly in last few years which discriminate stego images from non-stego images. The performance of a Steganalysis depends not only on the choice of classifier but also on features that are used to represent the image. Features extracted from images may contain irrelevant and redundant features which makes them inefficient for machine learning. Relevant features not only decrease the processing time to train a classifier but also provide better generalization. In this paper, kullback divergence measure, chernoff distance measure and linear regression are used for relevant feature selection. The performance of steganalysis using different measures used for feature selection is compared and evaluated in terms of classification error and computation time of training classifier. Experimental results show that Linear regression measure used for feature selection outperforms other measures used for feature selection in terms of both classification error and compilation time.