پنهان شكني كور در تصاوير JPEG خاكستري

چكيده انگليسي :

96 Blind Steganalysis of Graylevel JPEG Images Amir Hossein Haji Abdolhamidi a hajiabdolhamidi@ec iut ac ir August 6 2011 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Dr Abdolreza Mirzaei mirzaei@cc iut ac irAbstractHuman beings have always wanted to communicate in secure ways For many reasons such as transmitting adata on the network or sending a message to a friend exchanging the data in secret is essential There aremany solutions to keep the data secure among which an important one is cryptography Sensitiveinformation has been protected using encryption from since many years ago In cryptography powerfulmathematics is used to change plaintext into an unreadable coded text that is sent over a channel to thereceiver The other method of communication called steganography offers data protection in a differentmanner Steganography is the science of data communication securely in a digital media such as image audio and video files among them image files are most common especially JPEG images because theseimages are very popular and widely used Along with the rapid growth and improvements of steganographictechniques the attentions are turned into detection of these secret messages as well This process is calledSteganalysis Usually steganalysis methods are divided into two categories specific steganalysis and blindsteganalysis The specific steganalysis can recover the secret message or even estimate the embedding ratiowith the knowledge of the steganographic algorithm However implementation of this steganalysis method isvery hard because detection of steganography algorithm is difficult for steganalyzers or even the embeddingalgorithm may be unknown But blind steganalysis can detect the secret message independent of theembedding algorithm and is commonly used in many applications Blind steganalysis is done in two phases First the features that are changed with data embedding areextracted from images Then a classifier is built to train with these features and distinguish between coverand stego images Most of the researches are done with these two steps to reach better detection accuracy inblind steganalysis Some researchers tried to extract features that have a better result in steganalysis and theothers focused to implement a new classifier to improve the detection ratio Accordingly the currentclassifiers have acceptable performance to classify images as such feature extraction methods are moresignificant Hence many of the current blind steganalysis methods are based on extracting features which are liable tochange under data embedding These methods have various performances in steganographic algorithms Hence it is impossible to choose a specific feature with same performance on all steganographic algorithms Moreover blind steganalysis methods do not take advantage of the content diversity of images and theclassifier in these systems is trained with a huge image set with different contents In this thesis a newmethod is proposed focusing on image contents and the image set is clustered based on this diversity Hencethe training images that have to be classified are reduced Also one or more feature set which have betterresult in steganalysis are selected In testing phase of this framework two methods are proposed In the firstmethod combination of classifiers in all clusters is investigated In the other method called classifierselection a specific classifier is selected to reach the higher detection ratio In experimental results some ofthe current methods are implemented and compared to the proposed method in terms of detection accuracy In addition some statistical tests are performed in order to highlight the significant differences of the results KeywordsBlind Steganalysis Data Clustering Feature Extraction Classification Feature Selection

حاجي عبدالحميدي، اميرحسين

پنهان شكني كور در تصاوير JPEG خاكستري

خوشه بندي , استخراج ويژگي ها , دسته بندي