27.06.2013 Views

6th European Conference - Academic Conferences

6th European Conference - Academic Conferences

6th European Conference - Academic Conferences

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Detection of YASS Using Calibration by Motion Estimation<br />

Kesav Kancherla and Srinivas Mukkamala<br />

(ICASA) / (Canes) / New Mexico Institute of Mining and Technology USA<br />

kancherla@cs.nmt.edu<br />

srinivas@cs.nmt.edu<br />

Abstract: Through this paper we propose a new approach to thwart defects of current blind steganalysis<br />

methods. “Yet Another Steganographic Scheme” (YASS) is a robust steganographic scheme that embeds data in<br />

random locations based on a secret key. Due to this randomization the current steganalysis schemes such as<br />

self calibration methods do not detect YASS. In this work, we present a new calibration method using Motion<br />

Estimation and extract higher order features. In our methodology motion estimation technique is applied on an<br />

image, to estimate its actual image. We assume that the estimated image captures the features of the actual<br />

image, due to spatial redundancy in the images. We extract two sets of features; DCT based features from DCT<br />

domain and Markov model based features from spatial domain, and apply Support Vector Machines (SVMs) on<br />

these feature sets. Our approach against YASS using different block sizes (9, 10, 12, and 14), compression rates<br />

(50-50, 50/75, and 75/75) and coefficients used for embedding data (12 and 19) obtained an accuracy of about<br />

95%, even for bigger block lengths and low embedding rates. This methodology can be used as blind<br />

steganalysis technique, as detection is based on modification of an image rather than steganographic scheme.<br />

Keywords: blind steganalysis, Discrete Cosine Transform (DCT), motion estimation, steganalysis, Support<br />

Vector Machines (SVM)<br />

1. Introduction<br />

Steganography is the science of embedding data into cover object in covert communication. The rapid<br />

growth in internet and digital media causes an increasing threat of using steganography for covert<br />

communication. Steganographic images are not perceivable to human eye but embedding data into<br />

images change the statistics of images. The goal of a steganalyst is to use these statistical changes<br />

to detect the presence of any hidden message.<br />

Fridrich used second order statistics in her research of self-calibration method for blind Steganalysis<br />

(Fridrich, 2004: 67-81). In self-calibration technique, a given image is first decompressed and few<br />

rows and columns are cropped. The cropped image is recompressed using the same quality factor,<br />

and difference between the features extracted from actual image and cropped image is used to detect<br />

steganograms. To detect well-known steganographic schemes like Outguess, F5 and Model Based<br />

steganography schemes (Provos, 2001: 24; Westfeld, 2001: 289-302; Sallee, 2005: 167-190); Farid<br />

proposed the use of wavelet based features for JPEG Steganalysis (Lyu and Farid, 2002: 340-354),<br />

Shi proposed the use of transition matrix as features for detecting steganograms (Shi et al, 2006: 249-<br />

264), Fridrich used merged Discrete Cosine Transform (DCT) and Markov features for implementing a<br />

multi-class JPEG steganalysis classification (Pevny and Fridrich, 2007: 1-13) and Chen proposed<br />

Markov based features using intra-block and inter-block correlation of DCT coefficients (Chen and<br />

Shi, 2008: 3029-3032).<br />

Outguess embeds data by replacing least significant bit and preserves the first order statistics by<br />

performing additional changes, F5 algorithm uses matrix embedding to reduce the number of changes<br />

needed to embed data. And Model-based steganography tries to preserve histograms of individual<br />

AC DCT models after embedding the data. However the current steganalysis techniques can detect<br />

these steganography methods. “Yet Another Steganography Scheme” (YASS) by (Solanki, Sarkar<br />

and Manjunath, 2007: 16-31) is a new steganography scheme that resists the above steganalysis<br />

methods. YASS embeds data at random locations and uses Quantization Index Modulation (QIM) to<br />

increase robustness of data. Even though it cannot be detected using current self-calibration<br />

methods, embedding data still changes the statistical properties of image.<br />

In (Li, Shi and Huang, 2008: 139-148), the authors present a targeted attack on YASS. They showed<br />

that due to QIM embedding scheme used in YASS, there is an increase in number of zero DCT<br />

coefficients in stego image. Thus there is a notable difference between statistics of embedded block<br />

and the actual block. They also identified the fact that embedding is not random enough for detection<br />

of YASS. However this approach does not work when there are modifications in algorithm. In the<br />

method proposed by (Kodovský, Pevný and Fridrich, 2010: 1-11), the authors used various well<br />

known steganalysis methods for detection of YASS. They used Subtractive Pixel Adjacency Model<br />

143

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!