Your SlideShare is downloading. ×
0
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
PPT
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

PPT

625

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
625
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
13
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. By -chandana kaza
  • 2. <ul><li>Steganography </li></ul><ul><li>Steganalysis </li></ul><ul><li>Conventional methods </li></ul><ul><li>Machine learning based steganalysis </li></ul><ul><li>Experiments and results </li></ul>
  • 3. <ul><li>Transmit secret messages. </li></ul><ul><li>To make transferred secret messages undetectable. </li></ul><ul><li>Embed messages in such a way so as not be detected. </li></ul>
  • 4. &nbsp;
  • 5. <ul><li>Passive warden-examines and determines whether the message contains hidden message. </li></ul><ul><li>Active warden-alters the message, even though there is no trace of secret message. </li></ul>
  • 6. &nbsp;
  • 7. &nbsp;
  • 8. <ul><li>Not suited </li></ul><ul><ul><li>Images with low number of colors </li></ul></ul><ul><ul><li>Images with unique semantic content </li></ul></ul><ul><li>Best suited </li></ul><ul><ul><li>Gray scale images </li></ul></ul><ul><ul><li>Uncompressed scans of photographs </li></ul></ul><ul><ul><li>Images captured by digital camera </li></ul></ul>
  • 9. <ul><li>Simple and straight forward </li></ul><ul><li>Embed message into the least significant bit plane. </li></ul><ul><li>Difficult to be found by human eye. </li></ul>
  • 10. &nbsp;
  • 11. <ul><li>Cover image </li></ul><ul><li>Stego image </li></ul>
  • 12. <ul><li>To detect the existence of steganography </li></ul><ul><li>Estimate the message length </li></ul><ul><li>Extract hidden information </li></ul><ul><li>Achieved by exploiting differences between files. </li></ul>
  • 13. <ul><li>2 types of LSB embedding </li></ul><ul><ul><li>Sequential </li></ul></ul><ul><ul><li>Non-sequential </li></ul></ul><ul><ul><li>Classifying of steganalysis techniques </li></ul></ul><ul><ul><li>Instance based </li></ul></ul><ul><ul><li>Non-instance based </li></ul></ul>
  • 14. &nbsp;
  • 15. <ul><li>The  2 Method [Pfitzmann and Westfeld] </li></ul><ul><li>Splits images into segments </li></ul><ul><li>Calculate  2 co-efficients for every segments </li></ul><ul><li>Decide whether there is hidden message. </li></ul><ul><li>Suitable for sequentisl LSB </li></ul>
  • 16. <ul><li>If embedded message bit and original bit are different then, flip bit. </li></ul><ul><li>Technique </li></ul><ul><ul><li>Let pixel value=j </li></ul></ul><ul><ul><li>If j=2i, after flip j=2i+1 </li></ul></ul><ul><ul><li>If j=2i+1, after flip j=2i </li></ul></ul>
  • 17. <ul><li>Combines 2 pixel values 2i and 2i+1 together as a pair, and the two values differ in the lowest bit. </li></ul>
  • 18. &nbsp;
  • 19. <ul><li>Where k is the total number of all possible pixels. </li></ul><ul><li>If p is close to 1 then image is embedded. </li></ul>
  • 20. <ul><li>Suitable for any type of situation. </li></ul><ul><li>Exploits spatial correlation in images. </li></ul><ul><li>Based on analyzing how the number of R and S groups changes with the increased message length embedded in LSB plane. </li></ul>
  • 21. <ul><li>Technique </li></ul><ul><ul><li>Consider a MXN image with pixel values from set p. </li></ul></ul><ul><ul><li>For gray scale p={0……255} </li></ul></ul><ul><ul><li>Divide the image into disjoint groups of n adjacent pixels. </li></ul></ul><ul><ul><li>Define a discrimination function f(xl . . . . . xn)ER that assigns a real number to each pixel group </li></ul></ul>
  • 22. <ul><li>Purpose of the function is to quantify the smoothness or “regularity” of the group of pixels. </li></ul><ul><li>The noisier the group of pixels G=(x1….xn), the larger the value of the discrimination function. </li></ul>
  • 23. <ul><li>Define an invertible operation F on p called flipping. </li></ul><ul><li>LSB Flipping F 1 :0&lt;-&gt;1,2&lt;-&gt;3,….,254&lt;-&gt;255 </li></ul><ul><li>Shifted Lsb Flipping F -1 :-1&lt;-&gt;0,1&lt;-&gt;2,….,255&lt;-&gt;256 </li></ul>
  • 24. <ul><li>Depending on the above operation we define 3 pixel groups </li></ul><ul><li>Where F(G)=(F(x1),….f(xn)) </li></ul><ul><li>The flipping function can be captured by a mask M, which is a n-tuple with values -1,0,1. </li></ul>
  • 25. <ul><li>Number of R groups for mask M:RM </li></ul><ul><li>Number of singular groups:SM </li></ul><ul><li>Statistical hypothesis </li></ul><ul><ul><li>For typical image </li></ul></ul><ul><ul><li>This theory is violated after randomizing the LSB plane. </li></ul></ul>
  • 26. <ul><li>Randomization of the LSB plane forces </li></ul><ul><li>ie, the difference tend to become zero as the message length increases. </li></ul><ul><li>In the case of R -M and S -M the opposite happens. </li></ul>
  • 27. <ul><li>P is the length of message. </li></ul><ul><li>The initial measurement of R and S groups is R M( p/2), S M (p/2), R- M (p/2), and </li></ul><ul><li>S- M (p/2) </li></ul>
  • 28. <ul><li>The points RM(p/2), RM(1/2), RM(1-p/2) and SM(p/2), SM(1/2), SM(1-p/2) determine 2 parabolas. </li></ul><ul><li>Now calculate the root of the quadratic equation </li></ul>
  • 29. &nbsp;
  • 30. &nbsp;
  • 31. <ul><li>Learning denotes changes in the system that enable the system to do the same task more effectively next time. </li></ul><ul><li>Ex: classify an object as an instance </li></ul>
  • 32. <ul><li>The conventional methods just used some hypothesis observed heuristically. </li></ul><ul><li>If they are differences between the real model and fixed model, they will fail. </li></ul><ul><li>ML is used to reduce the errors brought by fixed models. </li></ul>
  • 33. <ul><li>For hidden information detection simple classifiers are used. </li></ul><ul><li>Using machine learning the quality of classifiers will be improved and successively more stable performance can be acquired. </li></ul>
  • 34. <ul><li>Hidden information process is treated as classification process. </li></ul><ul><li>I/p-images </li></ul><ul><li>O/p-class labels </li></ul><ul><li>The data set is built based on the values in  2 and RS methods. </li></ul>
  • 35. &nbsp;
  • 36. &nbsp;
  • 37. <ul><li>Training set : A portion of data set used to fit(train) a model for prediction or classification of values that are known in the training set, but unknown in (future)data. </li></ul><ul><li>Test set : A set of data used only to assess the performance [generalization] of a fully-specified classifier. </li></ul><ul><li>Feature extraction : Transforming the input data into the set of features to reduce redundant information is called features extraction. </li></ul>
  • 38. <ul><li>24 bit color images are collected </li></ul><ul><li>Embed different length of messages into images </li></ul><ul><li>Extract features using different methods for sequential and non-sequential cases. </li></ul><ul><li>Perform preprocessing </li></ul><ul><li>Every image will result an instance represented by a set of features in the data set. </li></ul>
  • 39. <ul><li>Build the experiment platform on WEKA </li></ul><ul><li>Test results on different machine learning methods. </li></ul><ul><ul><li>Naïve Bayes </li></ul></ul><ul><ul><li>Bayes networks </li></ul></ul><ul><ul><li>Decision trees </li></ul></ul><ul><ul><li>KNN </li></ul></ul><ul><ul><li>SVM </li></ul></ul><ul><ul><li>Neural networks </li></ul></ul>
  • 40. <ul><li>Apply ML-based classifier on POV3 algorithm. </li></ul><ul><li>The LSB bit-plane is treated as sequential pixel samples and is split into 100 segments. </li></ul><ul><li>For every segment, the  2 probability for all the pixels from the first segment to current one is calculated. </li></ul><ul><li>We get 300 coefficients, simple  2 then make a decision according to a threshold. </li></ul><ul><li>But our method uses these as features and construct classifiers. </li></ul>
  • 41. &nbsp;
  • 42. &nbsp;
  • 43. <ul><li>Precision α 1/Image complexity </li></ul><ul><li>Precision α embed rate </li></ul>
  • 44. <ul><ul><li>Because most of the pictures are in high complexity level 2-4, so ML-based methods generally perform better than simple  2 . </li></ul></ul><ul><ul><li>Conclude that applying machine learning to  2 can effectively improve the accuracy </li></ul></ul><ul><ul><li>Classifier wrapped conventional steganalysis maybe a good solution to detect sequential LSB steganography. </li></ul></ul>
  • 45. <ul><li>We use RS approach in this case </li></ul><ul><li>The differences between R±M(p/2), and S±M(p/2), increase when message length p increases. </li></ul><ul><li>The features are calculated using the difference </li></ul>
  • 46. <ul><li>Direct difference is not used in order to reduce the bias between different images. </li></ul><ul><li>Test this feature based methods with ML techniques. </li></ul><ul><li>Main focus is on the change of embed rate, the difference between different intrinsic complexities. </li></ul>
  • 47. Precision (RMS) Embed 0.1 Embed 0.2 Embed 0.5 Embed 1.0 Embed All Mixed Naive Bayes 50.90%(0.59) 53.80%(0.57) 54.90%(0.45) 94.56%(0.31) 80.48%(0.37) Bayes Net 89.21%(0.27) 95.85%(0.17) 99.35%(0.07) 99.45%(0.07) 95.44%(0.18) kNN 92.16%(0.28) 97.55%(0.16) 99.45%(0.07) 99.70%(0.05) 96.38%(0.19) J48 94.11%(0.27) 98.05%(0.14) 99.40%(0.08) 99.65%(0.06) 97.56%(0.14) SMO 59.39%(0.64) 75.32%(0.50) 93.21%(0.26) 96.70%(0.18) 80.90%(0.44) BP 53.10%(0.50) 54.10%(0.50) 52.70%(0.50) 56.80%(0.50) 80.00%(0.40) Threshold RS 95.30% 98.75% 99.70% 87.11% 97.38%
  • 48. <ul><ul><li>From table, we can see that J48 performs best in mixed embed rate case, and can get nearly 98% accuracy at all embed levels. </li></ul></ul><ul><ul><li>We use only two features, this result is comparable to  2 case in sequential embedding and is better than threshold based RS can do. </li></ul></ul>
  • 49. <ul><li>1. http://www.cs.waikato.ac.nz/ml/weka/ . </li></ul><ul><li>2. http://www.outguess.org/ . </li></ul><ul><li>3. S. Antani, R. Kasturi, and R. Jain. A survey on the use of pattern recognition methods for abstraction,indexing and retrieval of images and video. Pattern Recognition, 35(4):945{965, 2002. </li></ul><ul><li>4. G. Berg, I. Davidson, M.-Y. Duan, and G. Paul. Searching for hidden messages: Automatic detection of steganography. In IAAI, pages 51-56, 2003. </li></ul><ul><li>5. M. Morkel. Steganography And Steganalysis, ICSA Research Group. University of Pretoria, South Africa. January 2005. </li></ul><ul><li>6. Y.-M. Di, H. Liu, A. Ramineni, and A. Sen. Detecting hidden information in images: A comparative study. In 2nd Workshop on Privacy Preserving Data Mining (PPDM), 2003. </li></ul><ul><li>7. S. Dumitrescu, X. Wu, and Z. Wang. Detection of lsb steganography via sample pair analysis. In Information Hiding 5th International Workshop IH 2002 Revised Papers, Lecture Notes in Computer Science vol. 2578, pages 355{372, 2003. </li></ul><ul><li>8. J. J. Fridrich. Feature-based steganalysis for jpeg images and its implications for future design of steganographic schemes. In Information Hiding 6 th International Workshop IH 2004 Revised Selected Papers, Lecture Notes in Computer Science vol. 3200, pages 67{81, 2004. </li></ul>
  • 50. Thank you

×