X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/canny.git/blobdiff_plain/af81455342fce2b4d7f96ecb4f1194635a5a13a4..refs/heads/master:/experiments.tex?ds=sidebyside

diff --git a/experiments.tex b/experiments.tex
index 5ba43df..aa49e84 100644
--- a/experiments.tex
+++ b/experiments.tex
@@ -1,5 +1,5 @@
 First of all, the whole code of STABYLO can be downloaded
-\footnote{\url{http://members.femto-st.fr/raphael-couturier/stabylo/}}.
+\footnote{\url{http://http://members.femto-st.fr/jf-couchot/en/stabylo}}.
 For all the experiments, the whole set of 10,000 images 
 of the BOSS contest~\cite{Boss10} database is taken. 
 In this set, each cover is a $512\times 512$
@@ -42,7 +42,7 @@ $$
 \hline
 \end{array}
 $$
-\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}
 
 
@@ -62,80 +62,80 @@ this cover with respect to this embedding rate.
 
  
 
-\subsection{Image quality}\label{sub:quality}
-The visual quality of the STABYLO scheme is evaluated in this section.
-For the sake of completeness, three metrics are computed in these experiments: 
-the Peak Signal to Noise Ratio (PSNR), 
-the PSNR-HVS-M family~\cite{psnrhvsm11}, 
-and 
-the weighted PSNR (wPSNR)~\cite{DBLP:conf/ih/PereiraVMMP01}.
-The first one is widely used but does not take into
-account the Human Visual System (HVS).
-The other ones have been designed to tackle this problem.
+% \subsection{Image quality}\label{sub:quality}
+% The visual quality of the STABYLO scheme is evaluated in this section.
+% For the sake of completeness, three metrics are computed in these experiments: 
+% the Peak Signal to Noise Ratio (PSNR), 
+% the PSNR-HVS-M family~\cite{psnrhvsm11}, 
+% and 
+% the weighted PSNR (wPSNR)~\cite{DBLP:conf/ih/PereiraVMMP01}.
+% The first one is widely used but does not take into
+% account the Human Visual System (HVS).
+% The other ones have been designed to tackle this problem.
 
-If we apply them on the running example with the Adaptive and STC strategies, 
-the PSNR, PSNR-HVS-M, and wPSNR values are respectively equal to 
-68.39, 79.85, and 89.71 for the stego Lena when $b$ is equal to 7.
-If $b$ is 6, these values are respectively equal to 
-65.43, 77.2, and 89.35.
+% If we apply them on the running example with the Adaptive and STC strategies, 
+% the PSNR, PSNR-HVS-M, and wPSNR values are respectively equal to 
+% 68.39, 79.85, and 89.71 for the stego Lena when $b$ is equal to 7.
+% If $b$ is 6, these values are respectively equal to 
+% 65.43, 77.2, and 89.35.
 
 
 
 
-\begin{table*}
-\begin{center}
-\begin{small}
-\setlength{\tabcolsep}{3pt}
-\begin{tabular}{|c|c|c||c|c|c|c|c|c|c|c|c|c|}
-\hline
-Schemes & \multicolumn{4}{|c|}{STABYLO} & \multicolumn{2}{|c|}{HUGO}& \multicolumn{2}{|c|}{EAISLSBMR} & \multicolumn{2}{|c|}{WOW} & \multicolumn{2}{|c|}{UNIWARD}\\
-\hline
-Embedding &   Fixed & \multicolumn{3}{|c|}{Adaptive (about 6.35\%)} &  Fixed &Adaptive & Fixed &Adaptive & Fixed &Adaptive & Fixed &Adaptive \\
-\hline
-Rate &   10\% &  + sample &  +STC(7) & +STC(6) &  10\%&$\approx$6.35\%& 10\%&$\approx$6.35\%& 10\%&$\approx$6.35\%& 10\%&$\approx$6.35\%\\ 
-\hline
-PSNR & 61.86 & 63.48 &  66.55  &  63.7  & 64.65 & {67.08} & 60.8 & 62.9&65.9 & 68.3 & 65.8 & 69.2\\ 
-\hline
-PSNR-HVS-M & 72.9 & 75.39 & 78.6  & 75.5  & 76.67 & {79.6} & 71.8  & 76.0 &
-76.7 & 80.35 & 77.6 & 81.2 \\
-\hline
-wPSNR & 77.47 & 80.59 & 86.43& 86.28  & 83.03 & {88.6} & 76.7 & 83& 83.8 & 90.4 & 85.2 & 91.9\\ 
-\hline
-\end{tabular}
-\end{small}
-\end{center}
-\caption{Quality measures of steganography approaches\label{table:quality}}
-\end{table*}
+% \begin{table*}
+% \begin{center}
+% \begin{small}
+% \setlength{\tabcolsep}{3pt}
+% \begin{tabular}{|c|c|c||c|c|c|c|c|c|c|c|c|c|}
+% \hline
+% Schemes & \multicolumn{4}{|c|}{STABYLO} & \multicolumn{2}{|c|}{HUGO}& \multicolumn{2}{|c|}{EAISLSBMR} & \multicolumn{2}{|c|}{WOW} & \multicolumn{2}{|c|}{UNIWARD}\\
+% \hline
+% Embedding &   Fixed & \multicolumn{3}{|c|}{Adaptive (about 6.35\%)} &  Fixed &Adaptive & Fixed &Adaptive & Fixed &Adaptive & Fixed &Adaptive \\
+% \hline
+% Rate &   10\% &  + sample &  +STC(7) & +STC(6) &  10\%&$\approx$6.35\%& 10\%&$\approx$6.35\%& 10\%&$\approx$6.35\%& 10\%&$\approx$6.35\%\\ 
+% \hline
+% PSNR & 61.86 & 63.48 &  66.55  &  63.7  & 64.65 & {67.08} & 60.8 & 62.9&65.9 & 68.3 & 65.8 & 69.2\\ 
+% \hline
+% PSNR-HVS-M & 72.9 & 75.39 & 78.6  & 75.5  & 76.67 & {79.6} & 71.8  & 76.0 &
+% 76.7 & 80.35 & 77.6 & 81.2 \\
+% \hline
+% wPSNR & 77.47 & 80.59 & 86.43& 86.28  & 83.03 & {88.6} & 76.7 & 83& 83.8 & 90.4 & 85.2 & 91.9\\ 
+% \hline
+% \end{tabular}
+% \end{small}
+% \end{center}
+% \caption{Quality measures of steganography approaches\label{table:quality}}
+% \end{table*}
 
 
 
-Results are summarized in Table~\ref{table:quality}.
-In this table, STC(7) stands for embedding data in the LSB whereas
-in STC(6), data are hidden in the last two  significant bits. 
+% Results are summarized in Table~\ref{table:quality}.
+% In this table, STC(7) stands for embedding data in the LSB whereas
+% in STC(6), data are hidden in the last two  significant bits. 
 
 
-Let us give an interpretation of these experiments.
-First of all, the Adaptive strategy produces images with lower distortion 
-than the images resulting from the 10\% fixed strategy.
-Numerical results are indeed always greater for the former strategy than 
-for the latter one.
-These results are not surprising since the Adaptive strategy aims at 
-embedding messages whose length is decided according to a higher threshold
-into the edge detection.  
+% Let us give an interpretation of these experiments.
+% First of all, the Adaptive strategy produces images with lower distortion 
+% than the images resulting from the 10\% fixed strategy.
+% Numerical results are indeed always greater for the former strategy than 
+% for the latter one.
+% These results are not surprising since the Adaptive strategy aims at 
+% embedding messages whose length is decided according to a higher threshold
+% into the edge detection.  
 
 
-If we combine Adaptive and STC strategies 
-the STABYLO scheme  provides images whose quality is higher than 
-the EAISLSBMR's one but lower than the quality of high complexity 
-schemes. Notice that the quality of the less respectful scheme (EAILSBMR) 
-is lower than 6\% than the one of the most one.
+% If we combine Adaptive and STC strategies 
+% the STABYLO scheme  provides images whose quality is higher than 
+% the EAISLSBMR's one but lower than the quality of high complexity 
+% schemes. Notice that the quality of the less respectful scheme (EAILSBMR) 
+% is lower than 6\% than the one of the most one.
 
 
-% Let us now compare the STABYLO approach with other edge based steganography
-% approaches, namely~\cite{DBLP:journals/eswa/ChenCL10,Chang20101286}.
-% These two schemes focus on increasing the
-% payload while the PSNR is acceptable, but do not 
-% give quality metrics for fixed embedding rates from a large base of images. 
+% % Let us now compare the STABYLO approach with other edge based steganography
+% % approaches, namely~\cite{DBLP:journals/eswa/ChenCL10,Chang20101286}.
+% % These two schemes focus on increasing the
+% % payload while the PSNR is acceptable, but do not 
+% % give quality metrics for fixed embedding rates from a large base of images. 
 
 
 
@@ -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.
-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 
@@ -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}
-\caption{Steganalysing STABYLO\label{table:steganalyse}} 
+\caption{Steganalysing STABYLO\label{table:steganalyse}.} 
 \end{table*}
 
 
@@ -205,11 +208,29 @@ as far as we know.
 However by combining Adaptive and STC strategies
 our approach obtains similar results than the ones of these schemes.
 
-Compared to EAILSBMR, we obtain better results when the strategy is 
+Compared to EAILSBMR, we obtain similar
+results when the strategy is 
 Adaptive. 
-However due to its 
-huge number of integration features, it is not lightweight.
+However due to its huge number of integration features, it is not lightweight.
 
 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}