quelques modifs

[canny.git] / experiments.tex

diff --git a/experiments.tex b/experiments.tex
index 22900c0ec35afc7f4a5dc8937e27b2da0e0110f7..aa49e840c172619692b4e864b8cfaf195acf08d8 100644 (file)
--- a/experiments.tex
+++ b/experiments.tex
@@ -42,7 +42,7 @@ $$
 \hline
 \end{array}
 $$
 \hline
 \end{array}
 $$

-\caption{Matrix Generator for $\hat{H}$ in STC}\label{table:matrices:H}
+\caption{Matrix Generator for $\hat{H}$ in STC.}\label{table:matrices:H}

 \end{table}
 
 
 \end{table}
 
 


@@ -150,9 +150,12 @@ The steganalysis quality of our approach has been evaluated through the % two
 Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalyser.
 Its  particularization to spatial domain is 
 considered as state of the art steganalysers.
 Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalyser.
 Its  particularization to spatial domain is 
 considered as state of the art steganalysers.

-Firstly, a space 
-of 686 co-occurrence and Markov features is extracted from the 
-set of cover images and the set of training images. Next a small 
+Features that are embedded into this steganalysis process 
+are CCPEV and SPAM features as described 
+in~\cite{DBLP:dblp_conf/mediaforensics/KodovskyPF10}.
+They  are extracted from the 
+set of cover images and the set of training images.
+Next a small 

 set of weak classifiers is randomly built,
 each one working on a subspace of all the features.
 The final classifier is constructed by a majority voting 
 set of weak classifiers is randomly built,
 each one working on a subspace of all the features.
 The final classifier is constructed by a majority voting 


@@ -190,7 +193,7 @@ Ensemble Classifier & 0.35 & 0.44 & 0.47 & 0.47     & 0.48 &  0.49  &  0.43  & 0
 \end{tabular}
 \end{small}
 \end{center}
 \end{tabular}
 \end{small}
 \end{center}

-\caption{Steganalysing STABYLO\label{table:steganalyse}} 
+\caption{Steganalysing STABYLO\label{table:steganalyse}.} 

 \end{table*}
 
 
 \end{table*}
 
 


@@ -212,4 +215,22 @@ However due to its huge number of integration features, it is not lightweight.
 
 All these numerical experiments confirm 
 the objective presented in the motivations:
 
 All these numerical experiments confirm 
 the objective presented in the motivations:

-providing an efficient steganography approach in a lightweight manner.
+providing an efficient steganography approach in a lightweight manner
+for small payload.
+
+In Figure~\ref{fig:error}, 
+Ensemble Classifier has been used with all the previous 
+steganographic schemes with 4 different payloads.
+It can be observed that face to high values of payload, 
+STABYLO is definitely not secure enough.
+However thanks to an efficient very low-complexity (Fig.\ref{fig:compared}), 
+we argue that the user should embed tiny messages in many images 
+than a larger message in only one image.
+\begin{figure}
+\begin{center}
+\includegraphics[scale=0.5]{error}
+\end{center}
+\caption{Testing errors obtained by Ensemble classifier with 
+WOW/UNIWARD, HUGO, and STABYLO w.r.t. payload.}
+\label{fig:error} 
+\end{figure}