1 The flowcharts given in Fig.~\ref{fig:sch} summarize our steganography scheme denoted by
2 STABYLO, which stands for STeganography with cAnny, Bbs, binarY embedding at LOw cost.
3 What follows are successively details of the inner steps and flows inside
4 both the embedding stage (Fig.~\ref{fig:sch:emb})
5 and the extraction one (Fig.~\ref{fig:sch:ext}).
10 \subfloat[Data Embedding.]{
11 \begin{minipage}{0.49\textwidth}
13 %\includegraphics[width=5cm]{emb.pdf}
14 \includegraphics[width=5cm]{emb.ps}
19 \subfloat[Data Extraction.]{
20 \begin{minipage}{0.49\textwidth}
22 %\includegraphics[width=5cm]{rec.pdf}
23 \includegraphics[width=5cm]{rec.ps}
29 \caption{The STABYLO Scheme.}
36 \subsection{Data Embedding}
37 This section describes the main three steps of the STABYLO data embedding
42 \subsubsection{Edge-Based Image Steganography}
45 The edge-based image steganography schemes
46 already presented \cite{Luo:2010:EAI:1824719.1824720,DBLP:journals/eswa/ChenCL10} differ
47 in how carefully they select edge pixels, and
50 %Image Quality: Edge Image Steganography
51 %\JFC{Raphael, les fuzzy edge detection sont souvent utilisés.
52 % il faudrait comparer les approches en terme de nombre de bits retournés,
53 % en terme de complexité. } \cite{KF11}
54 %\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 :-)}
56 Many techniques have been proposed in the literature to detect
57 edges in images (whose noise has been initially reduced).
58 They can be separated in two categories: first and second order detection
59 methods on the one hand, and fuzzy detectors on the other hand~\cite{KF11}.
60 In first order methods like Sobel,
61 a first-order derivative (gradient magnitude, etc.) is computed
62 to search for local maxima, whereas in second order ones, zero crossings in a second-order derivative, like the Laplacian computed from the image,
63 are searched in order to find edges.
64 As for as fuzzy edge methods are concerned, they are obviously based on fuzzy logic to highlight
66 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.
69 %Of course, all the algorithms have advantages and drawbacks that depend on the
70 %motivations behind that edges detection. Unfortunately unless testing most of the
71 %algorithms, which would require many times, it is quite difficult to have an
72 %accurate idea on what would produce such algorithm compared to another.
75 As the Canny algorithm is well known and studied, fast, and implementable
76 on many kinds of architectures like FPGAs, smartphones, desktop machines, and
77 GPUs, we have chosen this edge detector for illustrative purpose.
78 Of course, other detectors like the fuzzy edge methods
79 deserve much further attention, which is why we intend
80 to investigate systematically all of these detectors in our next work.
81 %we do not pretend that this is the best solution.
83 In order to be able to compute the same set of edge pixels, we suggest to consider all the bits of the image (cover or stego) without the LSB. Thus, with an 8 bits image, only the 7 first bits are considered. In our flowcharts, this is represented by ``LSB(7 bits Edge Detection)''.
84 % First of all, let us discuss about compexity of edge detetction methods.
85 % Let then $M$ and $N$ be the dimension of the original image.
86 % According to~\cite{Hu:2007:HPE:1282866.1282944},
87 % even if the fuzzy logic based edge detection methods~\cite{Tyan1993}
88 % have promising results, its complexity is in $C_3 \times O(M \times N)$
89 % whereas the complexity on the Canny method~\cite{Canny:1986:CAE:11274.11275}
90 % is in $C_1 \times O(M \times N)$ where $C_1 < C_3$.
91 % \JFC{Verifier ceci...}
92 % In experiments detailled in this article, the Canny method has been retained
93 % but the whole approach can be updated to consider
94 % the fuzzy logic edge detector.
95 Next, following~\cite{Luo:2010:EAI:1824719.1824720}, our scheme automatically
96 modifies the Canny algorithm
97 parameters to get a sufficiently large set of edge bits: this
98 one is practically enlarged until its size is at least twice as large
99 as the size of the embedded message.
101 % Edge Based Image Steganography schemes
102 % already studied~\cite{Luo:2010:EAI:1824719.1824720,DBLP:journals/eswa/ChenCL10,DBLP:conf/ih/PevnyFB10} differ
103 % how they select edge pixels, and
104 % how they modify these ones.
106 % First of all, let us discuss about compexity of edge detetction methods.
107 % Let then $M$ and $N$ be the dimension of the original image.
108 % According to~\cite{Hu:2007:HPE:1282866.1282944},
109 % even if the fuzzy logic based edge detection methods~\cite{Tyan1993}
110 % have promising results, its complexity is in $C_3 \times O(M \times N)$
111 % whereas the complexity on the Canny method~\cite{Canny:1986:CAE:11274.11275}
112 % is in $C_1 \times O(M \times N)$ where $C_1 < C_3$.
113 % \JFC{Verifier ceci...}
114 % In experiments detailled in this article, the Canny method has been retained
115 % but the whole approach can be updated to consider
116 % the fuzzy logic edge detector.
118 % Next, following~\cite{Luo:2010:EAI:1824719.1824720}, our scheme automatically
119 % modifies Canny parameters to get a sufficiently large set of edge bits: this
120 % one is practically enlarged untill its size is at least twice as many larger
121 % than the size of embedded message.
124 \subsubsection{Security Considerations}
125 Among methods of message encryption/decryption
126 (see~\cite{DBLP:journals/ejisec/FontaineG07} for a survey)
127 we implement the Blum-Goldwasser cryptosystem~\cite{Blum:1985:EPP:19478.19501}
128 that is based on the Blum Blum Shub~\cite{DBLP:conf/crypto/ShubBB82} pseudorandom number generator (PRNG)
129 for security reasons.
130 It has been indeed proven~\cite{DBLP:conf/crypto/ShubBB82} that this PRNG
131 has the property of cryptographical security, \textit{i.e.},
132 for any sequence of $L$ output bits $x_i$, $x_{i+1}$, \ldots, $x_{i+L-1}$,
133 there is no algorithm, whose time complexity is polynomial in $L$, and
134 which allows to find $x_{i-1}$ and $x_{i+L}$ with a probability greater
136 Equivalent formulations of such a property can
137 be found. They all lead to the fact that,
138 even if the encrypted message is extracted,
139 it is impossible to retrieve the original one in
143 %%RAPH: paragraphe en double :-)
145 %% \subsubsection{Security Considerations}
146 %% Among methods of message encryption/decryption
147 %% (see~\cite{DBLP:journals/ejisec/FontaineG07} for a survey)
148 %% we implement the Blum-Goldwasser cryptosystem~\cite{Blum:1985:EPP:19478.19501}
149 %% which is based on the Blum Blum Shub~\cite{DBLP:conf/crypto/ShubBB82} Pseudo Random Number Generator (PRNG)
150 %% for security reasons.
151 %% It has been indeed proven~\cite{DBLP:conf/crypto/ShubBB82} that this PRNG
152 %% has the cryptographically security property, \textit{i.e.},
153 %% for any sequence $L$ of output bits $x_i$, $x_{i+1}$, \ldots, $x_{i+L-1}$,
154 %% there is no algorithm, whose time complexity is polynomial in $L$, and
155 %% which allows to find $x_{i-1}$ and $x_{i+L}$ with a probability greater
157 %% Thus, even if the encrypted message would be extracted,
158 %% it would thus be not possible to retrieve the original one in a
165 \subsubsection{Minimizing Distortion with Syndrome-Treillis Codes}
169 \subsection{Data Extraction}
170 The message extraction summarized in Fig.~\ref{fig:sch:ext} follows data embedding
171 since there exists a reverse function for all its steps.
172 First of all, the same edge detection is applied (on the 7 first bits) to
174 which is sufficiently large with respect to the message size given as a key.
175 Then the STC reverse algorithm is applied to retrieve the encrypted message.
176 Finally, the Blum-Goldwasser decryption function is executed and the original
177 message is extracted.