Degree Type

Thesis

Date of Award

2011

Degree Name

Master of Science

Department

Electrical and Computer Engineering

First Advisor

Jennifer L. Davidson

Abstract

A Canvass steganalyzer for double-compressed JPEG images

Steganography is the practice of hiding a secret message in innocent objects such that the very existence of the message is undetectable. Steganalysis, on the other hand, deals with finding the presence of such hidden messages. `Canvass' is software developed to perform JPEG image steganalysis. This software uses pattern recognizer to classify unknown images into cover (innocent) or stego (containing hidden message). The pattern recognizer, a support vector machine, is trained using the underlying statistical information in the cover and stego images. Some of the popular steganographic algorithms produce double-compressed JPEG images. A blind steganalyzer built on the assumption that it will see only single-compressed images gives misleading results of classification for such images. The goal of the current work is to develop a double-compression detector for JPEG images that extends the existing Canvass software. We develop a double-compression detector based on Partially Ordered Markov Models (POMMs) that can act as a pre-classifier to the blind steganalyzer. We also use the patterns of relative histogram values of the quantized DCT coefficients for improved accuracy of detection. After detecting the double- compression, we carry out cover Vs. stego detection and primary quality factor estimation. We compare our double-compression detector with two other state-of-the-art detectors. Our detector is found to have better performance compared to the state-of-the-art detectors. The current work considers a limited set of quality factors for double-compression but this novel method for steganalysis of double-compressed data looks promising and could be generalized for any combination of primary and secondary quality factors.

Copyright Owner

Pooja S. Paranjape

Language

en

Date Available

2012-04-06

File Format

application/pdf

File Size

68 pages

Share

COinS