X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/canny.git/blobdiff_plain/5e6f9e49a26cec4970434722a595a7b44197120f..8a7410beb8e5821fa7764d6fdc6843effeed1e0e:/experiments.tex diff --git a/experiments.tex b/experiments.tex index cf8dfa6..aa49e84 100644 --- a/experiments.tex +++ b/experiments.tex @@ -1,17 +1,236 @@ -\subsection{Steganalysis} - - - -The quality of our approach has been evaluated through the two -AUMP~\cite{Fillatre:2012:ASL:2333143.2333587} -and Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalysers. -Both aims at detecting hidden bits in grayscale natural images and are -considered as the state of the art of steganalysers in spatial domain~\cite{FK12}. -The former approach is based on a simplified parametric model of natural images. -Parameters are firstly estimated and a adaptive Asymptotically Uniformly Most Powerful -(AUMP) test is designed (theoretically and practically) to check whether -a natural image has stego content or not. -In the latter, the authors show that the -machine learning step, (which is often -implemented as support vector machine) -can be a favourably executed thanks to an Ensemble Classifiers. +First of all, the whole code of STABYLO can be downloaded +\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 $512\times 512$ +grayscale digital image in a RAW format. +We restrict experiments to +this set of cover images since this paper is more focused on +the methodology than on benchmarks. + +We use the matrices $\hat{H}$ +generated by the integers given +in Table~\ref{table:matrices:H} +as introduced in~\cite{FillerJF11}, since these ones have experimentally +be proven to have the strongest modification efficiency. +For instance if the rate between the size of the message and the size of the +cover vector +is 1/4, each number in $\{81, 95, 107, 121\}$ is translated into a binary number +and each one constitutes thus a column of $\hat{H}$. + +\begin{table} +$$ +\begin{array}{|l|l|} +\hline +\textrm{Rate} & \textrm{Matrix generators} \\ +\hline +{1}/{2} & \{71,109\}\\ +\hline +{1}/{3} & \{95, 101, 121\}\\ +\hline +{1}/{4} & \{81, 95, 107, 121\}\\ +\hline +{1}/{5} & \{75, 95, 97, 105, 117\}\\ +\hline +{1}/{6} & \{73, 83, 95, 103, 109, 123\}\\ +\hline +{1}/{7} & \{69, 77, 93, 107, 111, 115, 121\}\\ +\hline +{1}/{8} & \{69, 79, 81, 89, 93, 99, 107, 119\}\\ +\hline +{1}/{9} & \{69, 79, 81, 89, 93, 99, 107, 119, 125\}\\ +\hline +\end{array} +$$ +\caption{Matrix Generator for $\hat{H}$ in STC.}\label{table:matrices:H} +\end{table} + + +Our approach is always compared to HUGO, to EAISLSBMR, to WOW and to UNIWARD +for the two strategies Fixed and Adaptive. +For the former one, the payload has been set to 10\%. +For the latter one, the Canny parameter $T$ has been set to 3. +When $b$ is 7, the average size of the message that can be embedded +is 16,445 bits, +that corresponds to an average payload of 6.35\%. +For each cover image the STABYLO's embedding rate with these two parameters +is memorized. +Next each steganographic scheme is executed to produce the stego content of +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. + +% 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*} + + + +% 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. + + +% 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. + + + + +\subsection{Steganalysis}\label{sub:steg} + + + +The steganalysis quality of our approach has been evaluated through the % two +% AUMP~\cite{Fillatre:2012:ASL:2333143.2333587} +% and +Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalyser. +Its particularization to spatial domain is +considered as state of the art steganalysers. +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 +between the decisions of these individual classifiers. + + +%The former approach is based on a simplified parametric model of natural images. +% Parameters are firstly estimated and an adaptive Asymptotically Uniformly Most Powerful +% (AUMP) test is designed (theoretically and practically), to check whether +% an image has stego content or not. +% This approach is dedicated to verify whether LSB has been modified or not. +% , the authors show that the +% machine learning step, which is often +% implemented as a support vector machine, +% can be favorably executed thanks to an ensemble classifier. + + +\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 & Adapt. & Fixed & Adapt. & Fixed & Adapt. & Fixed & Adapt. \\ +\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 +%AUMP & 0.22 & 0.33 & 0.39 & 0.45 & 0.50 & 0.50 & 0.49 & 0.50 \\ +%\hline +Ensemble Classifier & 0.35 & 0.44 & 0.47 & 0.47 & 0.48 & 0.49 & 0.43 & 0.47 & 0.48 & 0.49 & 0.46 & 0.49 \\ + +\hline +\end{tabular} +\end{small} +\end{center} +\caption{Steganalysing STABYLO\label{table:steganalyse}.} +\end{table*} + + +Results of average testing errors +are summarized in Table~\ref{table:steganalyse}. +First of all, STC outperforms the sample strategy %for % the two steganalysers + as +already noticed in the quality analysis presented in the previous section. +Next, our approach is more easily detectable than HUGO, +WOW and UNIWARD which are the most secure steganographic tool, +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 similar +results when the strategy is +Adaptive. +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 +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}