The detection of data hiding in RGB images using statistical steganalysis

Other Title(s)

كشف البيانات المخفية في صور RGB باستخدام تحليل غطاء الإخفاء الإحصائي

Dissertant

Rasul, Zayd Ibrahim Rasul

Thesis advisor

al-Jarrah, Muzaffar Munir

Comitee Members

Alya, Muhammad Ahmad
Kayid, Ahmad

University

Middle East University

Faculty

Faculty of Information Technology

Department

Computer Science Department

University Country

Jordan

Degree

Master

Degree Date

2017

English Abstract

Steganalysis, the science and technology of detecting the presence of hidden data inside digital media, is a counter measure against information hiding techniques that can be used for illegitimate purposes.

The work in this thesis presents a steganalysis model that uses statistical texture features and the machine learning approach to detect the presence of hidden data in RGB color images.

The work analyzes features of an RGB image as a composite unit, as well as analyzing individual color channels and dual combinations of the channels.

The feature set used in this study consists of 26 features per channel, which includes the Gray Level Co-Occurrence Matrix (GLCM) features of correlation, contrast, homogeneity and energy, calculated for full bytes, half-bytes, 3-bit and 2-bit fragments of individual channels, Entropy of full bytes and half bytes, skewness of full bytes and half bytes, and additional statistical features.

The features are applied to single channels, and the single channel features are merged into dual and three-channel image feature sets.

The main machine learning binary classifier that is selected for this work is the Support Vector Machine (SVM) algorithm.

The experimental work used two image datasets of 1500 BMP images each, for training and validation of the model, and an independent image dataset of 1000 uncompressed PNG images for testing purposes.

Stego image datasets were created from the clean images datasets, which were embedded with secret data using 2LSB and 4LSB steganography techniques.

The experimental results for the validation phase showed detection accuracy of 100% for the 4LSB RGB stego images, and 99.73% for the 2LSB RGB stego images.

Similar results were obtained, which shows the power of the SVM classifier in detecting pattern changes in stego images even when one channel is changed, individual channels (R, G, B) and dual channels (RG, RB, GB) were analyzed.

Also, when only one channel was embedded with data, which was the blue channel, the same results were obtained.

The testing phase analyzed 1000 PNG stego images, which confirmed results of the validation phase.

The Discriminant Analysis (DA) classifier was used for comparison with the SVM classifier, and the results showed that the SVM classifier gave higher detection accuracy.

MATLAB 2015a was used in the implementation of the image processing and classification parts of the proposed model

Main Subjects

Information Technology and Computer Science

Topics

• Cryptography

No. of Pages

77

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-762708