University of Konstanz
Graduiertenkolleg / PhD Program
Computer and Information Science

Guest Talks


JPEG Steganalysis Based on Optimized Feature Extraction and Hierarchical Decision Making


Mohammad Rezaei
Teheran, Iran

date & place

Tuesday, 05.10.2010, 14:15 h
Room Z 714


This paper presents a novel comprehensive steganalysis scheme for JPEG images. The optimized features are extracted and a hierarchical decision-making is used to classify input images into cover and stego groups. The features have high discrimination ability mainly because they have been defined based on a careful study of different steganographic algorithms and their effects on statistical characteristics of images. We will show that first-order statistics of DCT coefficients are often more successful in the detection of LSB replacement steganographic methods such as JSteg, OutGuess, JPHide&Seek and StegHide, while second-order statistics have better performance in LSB Matching approaches and some other methods like MB1, SSIS and PQ. A database including 16000 cover JPEG images with 8 different quality factors has been used in our experiments. Experiments indicate that the accuracy of our method is higher than some other state of the art steganalysis methods and it has more generality. For example, for JPEG quality factor of 77, the mean True Positive and False Positive rates for embedded images with 6 different steganographic methods (i.e. JSteg, OutGuess, F5, MB1, Sequential LSB Matching, Random LSB Matching) and payloads over 20%, are 74.45% and 3.83% respectively. As the payload increases to 40%, the mean True Positive rate reaches to 92.3%. As a classifier, SVM with RBF kernel is used.