1 This section first presents the embedding scheme through its
2 four main steps: the data encryption (Sect.~\ref{sub:bbs}),
3 the cover pixel selection (Sect.~\ref{sub:edge}),
4 the adaptive payload considerations (Sect.~\ref{sub:adaptive}),
5 and how the distortion has been minimized (Sect.~\ref{sub:stc}).
6 The message extraction is finally presented (Sect.\ref{sub:extract}) and a running example ends this section (Sect.~\ref{sub:xpl}).
9 The flowcharts given in Fig.~\ref{fig:sch}
10 summarize our steganography scheme denoted by
11 STABYLO, which stands for STeganography with cAnny, Bbs, binarY embedding at LOw cost.
12 What follows are successively details of the inner steps and flows inside
13 both the embedding stage (Fig.~\ref{fig:sch:emb})
14 and the extraction one (Fig.~\ref{fig:sch:ext}).
15 Let us first focus on the data embedding.
19 \subfloat[Data Embedding.]{
20 \begin{minipage}{0.49\textwidth}
22 %\includegraphics[width=5cm]{emb.pdf}
23 \includegraphics[scale=0.5]{emb.ps}
28 \subfloat[Data Extraction.]{
29 \begin{minipage}{0.49\textwidth}
31 %\includegraphics[width=5cm]{rec.pdf}
32 \includegraphics[scale=0.5]{rec.ps}
38 \caption{The STABYLO Scheme.}
49 \subsection{Security Considerations}\label{sub:bbs}
50 Among methods of message encryption/decryption
51 (see~\cite{DBLP:journals/ejisec/FontaineG07} for a survey)
52 we implement the Blum-Goldwasser cryptosystem~\cite{Blum:1985:EPP:19478.19501}
53 that is based on the Blum Blum Shub~\cite{DBLP:conf/crypto/ShubBB82}
54 pseudorandom number generator (PRNG) and the
56 It has been indeed proven~\cite{DBLP:conf/crypto/ShubBB82} that this PRNG
57 has the property of cryptographical security, \textit{i.e.},
58 for any sequence of $L$ output bits $x_i$, $x_{i+1}$, \ldots, $x_{i+L-1}$,
59 there is no algorithm, whose time complexity is polynomial in $L$, and
60 which allows to find $x_{i-1}$ and $x_{i+L}$ with a probability greater
62 Equivalent formulations of such a property can
63 be found. They all lead to the fact that,
64 even if the encrypted message is extracted,
65 it is impossible to retrieve the original one in
68 Starting thus with a key $k$ and the message \textit{mess} to hide,
69 this step computes a message $m$, which is the encrypted version of \textit{mess}.
72 \subsection{Edge-Based Image Steganography}\label{sub:edge}
77 already presented \cite{Luo:2010:EAI:1824719.1824720,DBLP:journals/eswa/ChenCL10} differ
78 in how carefully they select edge pixels, and
81 %Image Quality: Edge Image Steganography
82 %\JFC{Raphael, les fuzzy edge detection sont souvent utilisés.
83 % il faudrait comparer les approches en terme de nombre de bits retournés,
84 % en terme de complexité. } \cite{KF11}
85 %\RC{Ben, à voir car on peut choisir le nombre de pixel avec Canny. Supposons que les fuzzy edge soient retourne un peu plus de points, on sera probablement plus détectable... Finalement on devrait surement vendre notre truc en : on a choisi cet algo car il est performant en vitesse/qualité. Mais on peut aussi en utilisé d'autres :-)}
87 Many techniques have been proposed in the literature to detect
88 edges in images (whose noise has been initially reduced).
89 They can be separated in two categories: first and second order detection
90 methods on the one hand, and fuzzy detectors on the other hand~\cite{KF11}.
91 In first order methods like Sobel, Canny~\cite{Canny:1986:CAE:11274.11275}, \ldots,
92 a first-order derivative (gradient magnitude, etc.) is computed
93 to search for local maxima, whereas in second order ones, zero crossings in a second-order derivative, like the Laplacian computed from the image,
94 are searched in order to find edges.
95 As for as fuzzy edge methods are concerned, they are obviously based on fuzzy logic to highlight edges.
97 Canny filters, on their parts, are an old family of algorithms still remaining a state-of-the-art edge detector. They can be well approximated by first-order derivatives of Gaussians.
98 As the Canny algorithm is well known and studied, fast, and implementable
99 on many kinds of architectures like FPGAs, smartphones, desktop machines, and
100 GPUs, we have chosen this edge detector for illustrative purpose.
102 \JFC{il faudrait comparer les complexites des algo fuzy and canny}
105 This edge detection is applied on a filtered version of the image given
107 More precisely, only $b$ most
108 significant bits are concerned by this step, where
109 the parameter $b$ is practically set with $6$ or $7$.
110 If set with the same value $b$, the edge detection returns thus the same
111 set of pixels for both the cover and the stego image.
112 In our flowcharts, this is represented by ``edgeDetection(b bits)''.
113 Then only the 2 LSBs of pixels in the set of edges are returned if $b$ is 6,
114 and the LSB of pixels if $b$ is 7.
120 Let $x$ be the sequence of these bits.
121 The next section section presentsd how our scheme
122 adapts when the size of $x$ is not sufficient for the message $m$ to embed.
130 \subsection{Adaptive Embedding Rate}\label{sub:adaptive}
131 Two strategies have been developed in our scheme,
132 depending on the embedding rate that is either \emph{adaptive} or \emph{fixed}.
133 In the former the embedding rate depends on the number of edge pixels.
134 The higher it is, the larger the message length that can be inserted is.
135 Practically, a set of edge pixels is computed according to the
136 Canny algorithm with an high threshold.
137 The message length is thus defined to be less than
138 half of this set cardinality.
139 If $x$ is then to short for $m$, the message is splitted into sufficient parts.
140 In the latter, the embedding rate is defined as a percentage between the
141 number of modified pixels and the length of the bit message.
142 This is the classical approach adopted in steganography.
143 Practically, the Canny algorithm generates
144 a set of edge pixels related to a threshold that is decreasing
145 until its cardinality
150 Two methods may further be applied to select bits that
152 The first one randomly chooses the subset of pixels to modify by
153 applying the BBS PRNG again. This method is further denoted as to \emph{sample}.
154 The second one is a direct application of the
155 STC algorithm~\cite{DBLP:journals/tifs/FillerJF11}.
156 It is further referred to as \emph{adaptive+STC} and is detailled in the nex section.
162 % First of all, let us discuss about compexity of edge detetction methods.
163 % Let then $M$ and $N$ be the dimension of the original image.
164 % According to~\cite{Hu:2007:HPE:1282866.1282944},
165 % even if the fuzzy logic based edge detection methods~\cite{Tyan1993}
166 % have promising results, its complexity is in $C_3 \times O(M \times N)$
167 % whereas the complexity on the Canny method~\cite{Canny:1986:CAE:11274.11275}
168 % is in $C_1 \times O(M \times N)$ where $C_1 < C_3$.
169 % \JFC{Verifier ceci...}
170 % In experiments detailled in this article, the Canny method has been retained
171 % but the whole approach can be updated to consider
172 % the fuzzy logic edge detector.
180 \subsection{Minimizing Distortion with Syndrome-Treillis Codes}\label{sub:stc}
185 % Edge Based Image Steganography schemes
186 % already studied~\cite{Luo:2010:EAI:1824719.1824720,DBLP:journals/eswa/ChenCL10,DBLP:conf/ih/PevnyFB10} differ
187 % how they select edge pixels, and
188 % how they modify these ones.
190 % First of all, let us discuss about compexity of edge detetction methods.
191 % Let then $M$ and $N$ be the dimension of the original image.
192 % According to~\cite{Hu:2007:HPE:1282866.1282944},
193 % even if the fuzzy logic based edge detection methods~\cite{Tyan1993}
194 % have promising results, its complexity is in $C_3 \times O(M \times N)$
195 % whereas the complexity on the Canny method~\cite{Canny:1986:CAE:11274.11275}
196 % is in $C_1 \times O(M \times N)$ where $C_1 < C_3$.
197 % \JFC{Verifier ceci...}
198 % In experiments detailled in this article, the Canny method has been retained
199 % but the whole approach can be updated to consider
200 % the fuzzy logic edge detector.
202 % Next, following~\cite{Luo:2010:EAI:1824719.1824720}, our scheme automatically
203 % modifies Canny parameters to get a sufficiently large set of edge bits: this
204 % one is practically enlarged untill its size is at least twice as many larger
205 % than the size of embedded message.
209 %%RAPH: paragraphe en double :-)
214 \subsection{Data Extraction}\label{sub:extract}
215 The message extraction summarized in Fig.~\ref{fig:sch:ext} follows data embedding
216 since there exists a reverse function for all its steps.
217 First of all, the same edge detection is applied (on the 7 first bits) to
219 which is sufficiently large with respect to the message size given as a key.
220 Then the STC reverse algorithm is applied to retrieve the encrypted message.
221 Finally, the Blum-Goldwasser decryption function is executed and the original
222 message is extracted.
225 \subsection{Running Example}\label{sub:xpl}
226 In this example, the cover image is Lena
227 which is a 512*512 image with 256 grayscale levels.
228 The message is the poem Ulalume (E. A. Poe), which is constituted by 104 lines, 667
229 words, and 3754 characters, \textit{i.e.} 30032 bits.
230 Lena and the the first verses are given in Fig.~\ref{fig:lena}.
234 \begin{minipage}{0.4\linewidth}
235 \includegraphics[width=3cm]{Lena.eps}
237 \begin{minipage}{0.59\linewidth}
240 The skies they were ashen and sober;\linebreak
241 $~$ The leaves they were crisped and sere—\linebreak
242 $~$ The leaves they were withering and sere;\linebreak
243 It was night in the lonesome October\linebreak
244 $~$ Of my most immemorial year;\linebreak
245 It was hard by the dim lake of Auber,\linebreak
246 $~$ In the misty mid region of Weir—\linebreak
247 It was down by the dank tarn of Auber,\linebreak
248 $~$ In the ghoul-haunted woodland of Weir.
253 \caption{Cover and message examples} \label{fig:lena}
256 The edge detection returns 18641 and 18455 pixels when $b$ is
257 respectively 7 and 6. These edges are represented in Fig.~\ref{fig:edge}
262 \subfloat[$b$ is 7.]{
263 \begin{minipage}{0.49\linewidth}
265 %\includegraphics[width=5cm]{emb.pdf}
266 \includegraphics[scale=0.15]{edge7.eps}
271 \subfloat[$b$ is 6.]{
272 \begin{minipage}{0.49\linewidth}
274 %\includegraphics[width=5cm]{rec.pdf}
275 \includegraphics[scale=0.15]{edge6.eps}
281 \caption{Edge Detection wrt $b$.}
287 In the former configuration, only 9320 bits are available
288 for embeding whereas in the latter we have 9227.
289 In the both case, about the third part of the poem is hidden into the cover.
290 Results with \emph{adaptive+STC} strategy are presented in
291 Fig.~\ref{fig:lenastego}.
295 \subfloat[$b$ is 7.]{
296 \begin{minipage}{0.49\linewidth}
298 %\includegraphics[width=5cm]{emb.pdf}
299 \includegraphics[scale=0.15]{lena7.eps}
304 \subfloat[$b$ is 6.]{
305 \begin{minipage}{0.49\linewidth}
307 %\includegraphics[width=5cm]{rec.pdf}
308 \includegraphics[scale=0.15]{lena6.eps}
314 \caption{Stego Images wrt $b$.}
315 \label{fig:lenastego}
319 Finally, differences between the original cover and the stego images
320 are presented in Fig.~\ref{fig:lenadiff}. For each pixel pair of picel $X_{ij}$
321 $Y_{ij}$, $X$ and $Y$ being the cover and the stego content respectively,
322 The pixel value $V_{ij}$ of the difference is defined with the following map
326 0 & \textrm{if} & X_{ij} = Y_{ij} \\
327 75 & \textrm{if} & \abs{ (X_{ij} - Y_{ij})} = 1 \\
328 75 & \textrm{if} & \abs{ (X_{ij} - Y_{ij})} = 2 \\
329 225 & \textrm{if} & \abs{ (X_{ij} - Y_{ij})} = 1
333 This function allows to emphase differences between content.
337 \subfloat[$b$ is 7.]{
338 \begin{minipage}{0.49\linewidth}
340 %\includegraphics[width=5cm]{emb.pdf}
341 \includegraphics[scale=0.15]{diff7.eps}
346 \subfloat[$b$ is 6.]{
347 \begin{minipage}{0.49\linewidth}
349 %\includegraphics[width=5cm]{rec.pdf}
350 \includegraphics[scale=0.15]{diff6.eps}
356 \caption{Differences with Lena's Cover wrt $b$.}