1 For whole experiments, the whole set of 10000 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 benchmarking.
8 Our approach is always compared to Hugo~\cite{DBLP:conf/ih/PevnyFB10}
9 and to EAISLSBMR~\cite{Luo:2010:EAI:1824719.1824720}.
10 The former is the least detectable information hiding tool in spatial domain
11 and the later is the work that is close to ours, as far as we know.
15 First of all, in our experiments and with the adaptive scheme,
16 the average size of the message that can be embedded is 16,445 bits.
17 Its corresponds to an average payload of 6.35\%.
18 The two other tools will then be compared with this payload.
19 The Sections~\ref{sub:quality} and~\ref{sub:steg} respectively present
20 the quality analysis and the security of our scheme.
26 \subsection{Image quality}\label{sub:quality}
27 The visual quality of the STABYLO scheme is evaluated in this section.
28 For the sake of completeness, three metrics are computed in these experiments:
29 the Peak Signal to Noise Ratio (PSNR),
30 the PSNR-HVS-M family~\cite{psnrhvsm11},
31 %the BIQI~\cite{MB10},
33 the weighted PSNR (wPSNR)~\cite{DBLP:conf/ih/PereiraVMMP01}.
34 The first one is widely used but does not take into
35 account the Human Visual System (HVS).
36 The other ones have been designed to tackle this problem.
38 If we apply them on the running example,
39 the PSNR, PSNR-HVS-M, and wPSNR values are respectively equal to
40 68.39, 79.85, and 89.71 for the stego Lena when $b$ is equal to 7.
41 If $b$ is 6, these values are respectively equal to
42 65.43, 77.2, and 89.35.
49 \begin{tabular}{|c|c|c||c|c|c|c|c|c|}
51 Schemes & \multicolumn{4}{|c|}{STABYLO} & \multicolumn{2}{|c|}{HUGO}& \multicolumn{2}{|c|}{EAISLSBMR} \\
53 Embedding & Fixed & \multicolumn{3}{|c|}{Adaptive (about 6.35\%)} & \multicolumn{2}{|c|}{Fixed}& \multicolumn{2}{|c|}{Fixed} \\
55 Rate & 10\% & + sample & +STC(7) & +STC(6) & 10\%&6.35\%& 10\%&6.35\%\\
57 PSNR & 61.86 & 63.48 & 66.55 (\textbf{-0.8\%}) & 63.7 & 64.65 & {67.08} & 60.8 & 62.9\\
59 PSNR-HVS-M & 72.9 & 75.39 & 78.6 (\textbf{-0.8\%}) & 75.5 & 76.67 & {79.23} & 71.8 & 74.3\\
61 %BIQI & 28.3 & 28.28 & 28.4 & 28.28 & 28.28 & 28.2 & 28.2\\
63 wPSNR & 77.47 & 80.59 & 86.43(\textbf{-1.6\%})& 86.28 & 83.03 & {87.8} & 76.7 & 80.6\\
69 Variances given in bold font express the quality differences between
70 HUGO and STABYLO with STC+adaptive parameters.
74 \caption{Quality measures of steganography approaches\label{table:quality}}
79 Results are summarized in Table~\ref{table:quality}.
80 Let us give an interpretation of these experiments.
81 First of all, the adaptive strategy produces images with lower distortion
82 than the one of images resulting from the 10\% fixed strategy.
83 Numerical results are indeed always greater for the former strategy than
85 These results are not surprising since the adaptive strategy aims at
86 embedding messages whose length is decided according to an higher threshold
87 into the edge detection.
88 Let us focus on the quality of HUGO images: with a given fixed
89 embedding rate (10\%),
90 HUGO always produces images whose quality is higher than the STABYLO's one.
91 However our approach is always better than EAISLSBMR since this one may modify
92 the two least significant bits.
94 If we combine \emph{adaptive} and \emph{STC} strategies
95 (which leads to an average embedding rate equal to 6.35\%)
96 our approach provides equivalent metrics than HUGO.
97 In this column STC(7) stands for embedding data in the LSB whereas
98 in STC(6), data are hidden in the two last significant bits.
102 The quality variance between HUGO and STABYLO for these parameters
103 is given in bold font. It is always close to 1\% which confirms
104 the objective presented in the motivations:
105 providing an efficient steganography approach with a lightweight manner.
108 Let us now compare the STABYLO approach with other edge based steganography
109 approaches, namely~\cite{DBLP:journals/eswa/ChenCL10,Chang20101286}.
110 These two schemes focus on increasing the
111 payload while the PSNR is acceptable, but do not
112 give quality metrics for fixed embedding rates from a large base of images.
117 \subsection{Steganalysis}\label{sub:steg}
121 The steganalysis quality of our approach has been evaluated through the two
122 AUMP~\cite{Fillatre:2012:ASL:2333143.2333587}
123 and Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalysers.
124 Both aim at detecting hidden bits in grayscale natural images and are
125 considered as the state of the art of steganalysers in spatial domain~\cite{FK12}.
126 The former approach is based on a simplified parametric model of natural images.
127 Parameters are firstly estimated and an adaptive Asymptotically Uniformly Most Powerful
128 (AUMP) test is designed (theoretically and practically), to check whether
129 an image has stego content or not.
130 This approach is dedicated to verify whether LSB has been modified or not.
131 In the latter, the authors show that the
132 machine learning step, which is often
133 implemented as a support vector machine,
134 can be favorably executed thanks to an ensemble classifier.
140 \begin{tabular}{|c|c|c|c|c|c|c|c|c|}
142 Schemes & \multicolumn{4}{|c|}{STABYLO} & \multicolumn{2}{|c|}{HUGO}& \multicolumn{2}{|c|}{EAISLSBMR}\\
144 Embedding & Fixed & \multicolumn{3}{|c|}{Adaptive (about 6.35\%)} & \multicolumn{2}{|c|}{Fixed}& \multicolumn{2}{|c|}{Fixed} \\
146 Rate & 10\% & + sample & +STC(7) & +STC(6) & 10\%& 6.35\%& 10\%& 6.35\%\\
148 AUMP & 0.22 & 0.33 & 0.39 & 0.45 & 0.50 & 0.50 & 0.49 & 0.50 \\
150 Ensemble Classifier & 0.35 & 0.44 & 0.47 & 0.47 & 0.48 & 0.49 & 0.43 & 0.46 \\
156 \caption{Steganalysing STABYLO\label{table:steganalyse}}
160 Results are summarized in Table~\ref{table:steganalyse}.
161 First of all, STC outperforms the sample strategy for the two steganalysers, as
162 already noticed in the quality analysis presented in the previous section.
163 Next, our approach is more easily detectable than HUGO, which
164 is the most secure steganographic tool, as far as we know.
165 However by combining \emph{adaptive} and \emph{STC} strategies
166 our approach obtains similar results than HUGO ones.
169 huge number of features integration, it is not lightweight, which justifies
170 in the authors' opinion the consideration of the proposed method.