An Information Theoretic Image Steganalysis for LSB Steganography

Abstract

Steganography hides the data within a media file in an imperceptible way. Steganalysis exposes steganography by using detection measures. Traditionally, Steganalysis revealed steganography by targeting perceptible and statistical properties which results in developing secure steganography schemes. In this work, we target LSB image steganography by using entropy and joint entropy metrics for steganalysis. First, the Embedded image is processed for feature extraction then analyzed by entropy and joint entropy with their corresponding original image. Second, SVM and Ensemble classifiers are trained according to the analysis results. The decision of classifiers discriminates cover image from stego image. This scheme is further applied on attacked stego image for checking detection reliability. Performance evaluation of proposed scheme is conducted over grayscale image datasets. We analyzed LSB embedded images by Comparing information gain from entropy and joint entropy metrics. Results conclude that entropy of the suspected image is more preserving than joint entropy. As before histogram attack, detection rate with entropy metric is 70% and 98% with joint entropy metric. However after an attack, entropy metric ends with 30% detection rate while joint entropy metric gives 93% detection rate. Therefore, joint entropy proves to be better steganalysis measure with 93% detection accuracy and less false alarms with varying hiding ratio.

Image steganalysis with entropy hybridized with chaotic grasshopper optimizer

“…This section describes the steganalysis model with the proposed hybrid approach for reducing the dimensions of feature sets used in the detection process. The features based on entropy and mutual information have already been described in the previous section; see [7] for details of their implementation. The steganalysis process starts from the feature extraction step, but for experimental purposes, we include the embedding phase before the feature extraction phase, as shown in Fig.…”

“…In addition, we have found information-theoreticentropy and mutual informationfeature sets quite distinctive for pixel-based steganography [7]. Therefore, in this work, we use entropy and a combination of SPAM and SRM features from the image sets, the details of their computations are included, respectively in [7] and [8].…”

“…where P(u) and P(v) are marginal probability distributions of U and V variables. See [7] for details of the implementation.…”