1 For whole experiments, the whole set of 10,000 images
2 of the BOSS contest~\cite{Boss10} database is taken.
3 In this set, each cover is a $512\times 512$
4 grayscale digital image in a RAW format.
5 We restrict experiments to
6 this set of cover images since this paper is more focused on
7 the methodology than on benchmarking.
9 We use the matrices $\hat{H}$
10 generated by the integers given
11 in table~\ref{table:matrices:H}
12 as introduced in~\cite{FillerJF11}, since these ones have experimentally
13 be proven to have the best modification efficiency.
14 For instance if the rate between the size of the message and the size of the
16 is 1/4, each number in $\{81, 95, 107, 121\}$ is translated into a binary number
17 and each one consitutes thus a column of $\hat{H}$.
23 \textrm{Rate} & \textrm{Matrix generators} \\
25 {1}/{2} & \{71,109\}\\
27 {1}/{3} & \{95, 101, 121\}\\
29 {1}/{4} & \{81, 95, 107, 121\}\\
31 {1}/{5} & \{75, 95, 97, 105, 117\}\\
33 {1}/{6} & \{73, 83, 95, 103, 109, 123\}\\
35 {1}/{7} & \{69, 77, 93, 107, 111, 115, 121\}\\
37 {1}/{8} & \{69, 79, 81, 89, 93, 99, 107, 119\}\\
39 {1}/{9} & \{69, 79, 81, 89, 93, 99, 107, 119, 125\}\\
43 \caption{Matrix Generator for $\hat{H}$ in STC}\label{table:matrices:H}
47 Our approach is always compared to Hugo~\cite{DBLP:conf/ih/PevnyFB10}
48 and to EAISLSBMR~\cite{Luo:2010:EAI:1824719.1824720}.
49 The former is the least detectable information hiding tool in spatial domain
50 and the latter is the work that is the closest to ours, as far as we know.
54 First of all, in our experiments and with the adaptive scheme,
55 the average size of the message that can be embedded is 16,445 bits.
56 It corresponds to an average payload of 6.35\%.
57 The two other tools will then be compared with this payload.
58 Sections~\ref{sub:quality} and~\ref{sub:steg} respectively present
59 the quality analysis and the security of our scheme.
65 \subsection{Image quality}\label{sub:quality}
66 The visual quality of the STABYLO scheme is evaluated in this section.
67 For the sake of completeness, three metrics are computed in these experiments:
68 the Peak Signal to Noise Ratio (PSNR),
69 the PSNR-HVS-M family~\cite{psnrhvsm11},
70 %the BIQI~\cite{MB10},
72 the weighted PSNR (wPSNR)~\cite{DBLP:conf/ih/PereiraVMMP01}.
73 The first one is widely used but does not take into
74 account the Human Visual System (HVS).
75 The other ones have been designed to tackle this problem.
77 If we apply them on the running example,
78 the PSNR, PSNR-HVS-M, and wPSNR values are respectively equal to
79 68.39, 79.85, and 89.71 for the stego Lena when $b$ is equal to 7.
80 If $b$ is 6, these values are respectively equal to
81 65.43, 77.2, and 89.35.
88 \begin{tabular}{|c|c|c||c|c|c|c|c|c|}
90 Schemes & \multicolumn{4}{|c|}{STABYLO} & \multicolumn{2}{|c|}{HUGO}& \multicolumn{2}{|c|}{EAISLSBMR} \\
92 Embedding & Fixed & \multicolumn{3}{|c|}{Adaptive (about 6.35\%)} & \multicolumn{2}{|c|}{Fixed}& \multicolumn{2}{|c|}{Fixed} \\
94 Rate & 10\% & + sample & +STC(7) & +STC(6) & 10\%&6.35\%& 10\%&6.35\%\\
96 PSNR & 61.86 & 63.48 & 66.55 (\textbf{-0.8\%}) & 63.7 & 64.65 & {67.08} & 60.8 & 62.9\\
98 PSNR-HVS-M & 72.9 & 75.39 & 78.6 (\textbf{-0.8\%}) & 75.5 & 76.67 & {79.23} & 71.8 & 74.3\\
100 %BIQI & 28.3 & 28.28 & 28.4 & 28.28 & 28.28 & 28.2 & 28.2\\
102 wPSNR & 77.47 & 80.59 & 86.43(\textbf{-1.6\%})& 86.28 & 83.03 & {87.8} & 76.7 & 80.6\\
108 Variances given in bold font express the quality differences between
109 HUGO and STABYLO with STC+adaptive parameters.
113 \caption{Quality measures of steganography approaches\label{table:quality}}
118 Results are summarized in Table~\ref{table:quality}.
119 Let us give an interpretation of these experiments.
120 First of all, the adaptive strategy produces images with lower distortion
121 than the images resulting from the 10\% fixed strategy.
122 Numerical results are indeed always greater for the former strategy than
124 These results are not surprising since the adaptive strategy aims at
125 embedding messages whose length is decided according to an higher threshold
126 into the edge detection.
127 Let us focus on the quality of HUGO images: with a given fixed
128 embedding rate (10\%),
129 HUGO always produces images whose quality is higher than the STABYLO's one.
130 However our approach is always better than EAISLSBMR since this one may modify
131 the two least significant bits.
133 If we combine \emph{adaptive} and \emph{STC} strategies
134 (which leads to an average embedding rate equal to 6.35\%)
135 our approach provides metrics equivalent to those provided by HUGO.
136 In this column STC(7) stands for embedding data in the LSB whereas
137 in STC(6), data are hidden in the last two significant bits.
141 The quality variance between HUGO and STABYLO for these parameters
142 is given in bold font. It is always close to 1\% which confirms
143 the objective presented in the motivations:
144 providing an efficient steganography approach with a lightweight manner.
147 Let us now compare the STABYLO approach with other edge based steganography
148 approaches, namely~\cite{DBLP:journals/eswa/ChenCL10,Chang20101286}.
149 These two schemes focus on increasing the
150 payload while the PSNR is acceptable, but do not
151 give quality metrics for fixed embedding rates from a large base of images.
156 \subsection{Steganalysis}\label{sub:steg}
160 The steganalysis quality of our approach has been evaluated through the % two
161 % AUMP~\cite{Fillatre:2012:ASL:2333143.2333587}
163 Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalyser.
164 This approach aims at detecting hidden bits in grayscale natural
166 considered as state of the art steganalysers in the spatial domain~\cite{FK12}.
167 %The former approach is based on a simplified parametric model of natural images.
168 % Parameters are firstly estimated and an adaptive Asymptotically Uniformly Most Powerful
169 % (AUMP) test is designed (theoretically and practically), to check whether
170 % an image has stego content or not.
171 % This approach is dedicated to verify whether LSB has been modified or not.
172 % , the authors show that the
173 % machine learning step, which is often
174 % implemented as a support vector machine,
175 % can be favorably executed thanks to an ensemble classifier.
181 \begin{tabular}{|c|c|c|c|c|c|c|c|c|}
183 Schemes & \multicolumn{4}{|c|}{STABYLO} & \multicolumn{2}{|c|}{HUGO}& \multicolumn{2}{|c|}{EAISLSBMR}\\
185 Embedding & Fixed & \multicolumn{3}{|c|}{Adaptive (about 6.35\%)} & \multicolumn{2}{|c|}{Fixed}& \multicolumn{2}{|c|}{Fixed} \\
187 Rate & 10\% & + sample & +STC(7) & +STC(6) & 10\%& 6.35\%& 10\%& 6.35\%\\
189 %AUMP & 0.22 & 0.33 & 0.39 & 0.45 & 0.50 & 0.50 & 0.49 & 0.50 \\
191 Ensemble Classifier & 0.35 & 0.44 & 0.47 & 0.47 & 0.48 & 0.49 & 0.43 & 0.46 \\
197 \caption{Steganalysing STABYLO\label{table:steganalyse}}
201 Results are summarized in Table~\ref{table:steganalyse}.
202 First of all, STC outperforms the sample strategy %for % the two steganalysers
204 already noticed in the quality analysis presented in the previous section.
205 Next, our approach is more easily detectable than HUGO, which
206 is the most secure steganographic tool, as far as we know.
207 However by combining \emph{adaptive} and \emph{STC} strategies
208 our approach obtains similar results to HUGO ones.
213 Compared to EAILSBMR, we obtain better results when the strategy is
216 huge number of integration features, it is not lightweight, which justifies
217 in the authors' opinion the consideration of the proposed method.