Loading...

Structural and Algorithmic Analysis of Machine Learning for Steganalysis Based on Diversity and Size of Feature Space

Karimi, Saeed | 2014

782 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46190 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghaemmaghami, Shahrokh
  7. Abstract:
  8. In this project we proposed a new method for improving the detection abality of a steganalyser with a pre-processing on contents of an image. Steganalysis, using machine learning, is designing a classifier with two classes: Stego or Cover. This classifier should be trained with extracted features from signal. The result of the training procedure is a machine that decides a signal belongs to stego or cover class. The first step of steganalysis process is extraction of proper features from signal. Proper feature is a variable that represents all of the useful properties of signal. Second step of this process is classifying data to two class of stego and cover. Many algorithms are proposed for designing a classifier. Support vector machines introduces an apropriate method for classyfing of two classes problems. It is the most common method for classifying that maps data in feature co-ordinates. In trainig stage, classifier tries to make a hyperplane that separates tagged features of each classes. For improving the results and circumstances of steganalysis process, we can concentrate on two parts. First part is the extraction of proper features. This features should make differences between stego and cover objects. Second part is using an appropriate classifier. Experimental results of steganalysis processes shows that the trained model and true detection rate of steganalyzer is related to the samples used in training stage. Similarity between training and testing samples is a key factor in performance of steganalyzer, as well. In this project we proposed a method for improving performance of steganalyzer that points to three key factors: classifier, feature space and content of training data. This method is based on a two-level classifier. Second level of this classifier works like common classifiers. But the first level categorizes similar data. Some new features are proposed for categorizing. These features depend on the content of images and make two categories: "more" and "less". Therefore the main idea of this project is categorizing similar images and implementing steganalyzer on each category separately. Some proofs are proposed for demonstrating the efficiency of this idea. Experimental results show that the proposed method can improve true detection rate up to 3.5%. This method works on every steganography and steganalysis algorithms that their true detection rates are in a range that can be improved
  9. Keywords:
  10. Feature Space ; Classification ; Steganalysis ; Machine Learning ; Content Dependent ; Classifiers Combination ; Multilevel Classification ; True Detection Rate

 Digital Object List

 Bookmark

No TOC