![](bg6.png)
Table 4: Comparison of APCC results with other detection approaches (best results highlighted in bold font).
Classifier Detection Approach in APS Algorithm
Deogol Huang et al. Shen et al.
(Forrest, 2006) (Huang et al., 2008) (Shen and Zhao, 2010)
Acc AUC Acc AUC Acc AUC
MLP
Polak and Kotulski (Polak and Kotulski, 2010) 74.64% 0.78 47.32% 0.48 44.46% 0.33
Sedeeq et al.(Sedeeq et al., 2016) 75.36% 0.88 68.10% 0.84 61.37% 0.65
Jian-feng et al. (Jian-feng et al., 2014) 86.61% 0.93 78.93% 0.91 77.32% 0.88
APCC 93.39% 0.97 90.36% 0.94 93.57% 0.99
SVM
Polak and Kotulski (Polak and Kotulski, 2010) 61.96% 0.65 48.57% 0.50 47.32% 0.46
Sedeeq et al. (Sedeeq et al., 2016) 76.79% 0.77 74.70% 0.74 58.10% 0.59
Jian-feng et al.(Jian-feng et al., 2014) 88.04% 0.88 81.43% 0.82 75.54% 0.75
APCC 90.71% 0.90 93.21% 0.91 96.07% 0.95
NB
Polak and Kotulski (Polak and Kotulski, 2010) 70.63% 0.75 51.43% 0.47 51.61% 0.53
Sedeeq et al. (Sedeeq et al., 2016) 78.04% 0.89 74.70% 0.84 61.01% 0.65
Jian-feng et al.(Jian-feng et al., 2014) 91.96% 0.99 89.64% 0.91 79.82% 0.89
APCC 90.89% 0.98 88.75% 0.96 93.57% 0.99
algorithm used. Inspection of the table indicates that,
with respect to accuracy Acc, the APCC approach
produced best results in seven of the nine cases; and,
with respect to AUC, the best result in eight of the
nine cases. It is interesting to note that the dynamic
techniques of Polak and Kotulski, and Sedeeq et al.,
did not perform well. This is probably because neither
technique was well suited to usage in a static con-
text. It is also interesting to note that the technique
proposed by Jian-feng et al. worked well when us-
ing NB classification (Jian-feng et al, originally used
SVM classification to evaluate their approach).
6 CONCLUSION
In this paper a novel approach to detecting HTML
Attribute Permutation Steganography (APS) has been
suggested. The approach is founded on the usage of a
proposed Attribute Position Changes Count (APCC)
metric, the main contribution of the paper. This met-
ric offers the dual advantages that: (i) it serves to cap-
ture more detail concerning APS than methods that
use average statistical values and (ii) it can be readily
used to generate feature vectors with which to train an
APS classification model. The evaluation was con-
ducted by considering three alternative APS meth-
ods and three classifier generation paradigms (thus
three-by-three combinations) and in each case com-
paring the proposed APCC APS detection approach
with three alternative APS detection approaches from
the literature. In eight out of the nine cases the pro-
posed approach produced the best AUC value, and in
seven out of the nine cases the best accuracy ACC
value, thus indicating that the proposed approach can
be successfully employed to detect attribute permuta-
tion steganography. A deeper analysis is required in
order to understand cases when Jian-feng et al. (Jian-
feng et al., 2014) performs better than APCC. Further
work includes development and evaluation of novel
metrics and algorithms for HTML steganography de-
tection.
REFERENCES
Ayodele, T. (2010). Types of machine learning algorithms.
Demsar, J. (2006). Statistical comparisons of classifiers
over multiple data sets. Machine Learning Research,
pages 1–30.
Forrest, S. (2006). Introduction to deogol.
Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann,
P., and H., I. (2009). The weka data mining software:
An update; sigkdd explorations, volume 11, issue 1.
Huang, H., Zhong, S., and Sun, X. (2008). An algorithm of
webpage information hiding based on attributes per-
mutation. In 4th International Conference on Intelli-
gent Information Hiding and Multimedia Signal Pro-
cessing (IIH-MSP 2008), pages 257–260.
Jian-feng, W., Liu-sheng, H., Miao-miao, T., Zhi-li, C., and
Hai-bo, M. (2014). Detection of html steganography
based on statistics and svm. Journal of Chinese Com-
puter Systems, 35(6):1221–1225.
Polak, L. and Kotulski, Z. (2010). Sending hidden
data through www pages detection and prevention.
Engn.Trans., 58:75–89.
Scholkopf, B. and Smola, A. (2002). Learning with Ker-
nals. MIT Press.
Sedeeq, I., Coenen, F., and Lisitsa, A. (2016). A statistical
approach to the detection of html attribute permutation
steganography. Second ICISSP Italy.
Shen, D. and Zhao, H. (2010). A novel scheme of webpage
information hiding based on attributes. In Information
Theory and Information Security (ICITIS), 2010 IEEE
International Conference on, pages 1147–1150.
Smola, A. and Vishwanathan, S. (2008). Introduction to
Machine Learning. Cambridge University Press, The
Pitt Building, Trumpington Street, Cambridge, United
Kingdom.
Sui, X.-G. and Luo, H. (2004). A new steganography
method based on hypertext. In Radio Science Con-
ference, 2004. Proceedings. 2004 Asia-Pacific, pages
181–184.
Zhao, Q. and Lu, H. (2007). Pca-based web page water-
marking. Pattern Recogn., 40(4):1334–1341.
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
100