1 \subsection{Adaptive Embedding Rate}
3 Two strategies have been developed in our scheme with respect to the rate of
4 embedding which is either \emph{ adaptive} or \emph{fixed}.
6 In the former the embedding rate depends on the number of edge pixels.
7 The higher it is, the larger is the message length that can be considered.
8 Practically, a set of edge pixels is computed according to the
9 Canny algorithm with high threshold.
10 The message length is thus defined to be the half of this set cardinality.
11 The rate between available bits and bit message length is then more than two.This constraint is indeed induced by the fact that the efficiency
12 of the STC algorithm is unsatisfactory under that threshold.
15 In the latter, the embedding rate is defined as a percentage between the
16 number of the modified pixels and the length of the bit message.
17 This is the classical approach adopted in steganography.
18 Practically, the Canny algorithm generates a
19 a set of edge pixels with threshold that is decreasing until its cardinality
20 is sufficient. If the set cardinality is more than twice larger than the
21 bit message length an STC step is again applied.
22 Otherwise, pixels are randomly chosen from the set of pixels to build the
23 subset with a given size. The BBS PRNG is again applied there.
28 \subsection{Image Quality}
29 The visual quality of the STABYLO scheme is evaluated in this section.
30 Four metrics are computed in these experiments:
31 the Peak Signal to Noise Ratio (PSNR),
32 the PSNR-HVS-M family~\cite{PSECAL07,psnrhvsm11} ,
33 the BIQI~\cite{MB10,biqi11} and
34 the weighted PSNR (wPSNR)~\cite{DBLP:conf/ih/PereiraVMMP01}.
35 The first one is widely used but does not take into
36 account Human Visual System (HVS).
37 The other last ones have been designed to tackle this problem.
41 \begin{tabular}{|c|c|c|}
43 Embedding rate & Adaptive & 10 \% \\
45 PSNR & 66.55 & 61.86 \\
47 PSNR-HVS-M & 78.6 & 72.9 \\
51 wPSNR & 86.43& 77.47 \\
55 \caption{Quality measures of our steganography approach\label{table:quality}}
59 Let us compare the STABYLO approach with other edge based steganography
60 schemes with respect to the image quality.
61 First of all, wPSNR and PSNR of the Edge Adaptive
62 scheme detailed in~\cite{Luo:2010:EAI:1824719.1824720} are lower than ours.
63 Next both the approaches~\cite{DBLP:journals/eswa/ChenCL10,Chang20101286}
64 focus on increasing the payload while the PSNR is acceptable, but do not
65 give quality metrics for fixed embedding rate from a large base of images.
66 Our approach outperforms the former thanks to the introduction of the STC
70 \subsection{Steganalysis}
74 The quality of our approach has been evaluated through the two
75 AUMP~\cite{Fillatre:2012:ASL:2333143.2333587}
76 and Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalysers.
77 Both aims at detecting hidden bits in grayscale natural images and are
78 considered as the state of the art of steganalysers in spatial domain~\cite{FK12}.
79 The former approach is based on a simplified parametric model of natural images.
80 Parameters are firstly estimated and a adaptive Asymptotically Uniformly Most Powerful
81 (AUMP) test is designed (theoretically and practically) to check whether
82 a natural image has stego content or not.
83 In the latter, the authors show that the
84 machine learning step, (which is often
85 implemented as support vector machine)
86 can be a favourably executed thanks to an Ensemble Classifiers.
92 \begin{tabular}{|c|c|c|c|}
94 Schemes & \multicolumn{2}{|c|}{STABYLO} & HUGO\\
96 Embedding rate & Adaptive & 10 \% & 10 \%\\
98 AUMP & 0.39 & 0.22 & 0.50 \\
100 Ensemble Classifier & 0.47 & 0.35 & 0.48 \\
105 \caption{Steganalysing STABYLO\label{table:steganalyse}}
109 Results show that our approach is more easily detectable than HUGO which is
110 is the more secure steganography tool, as far we know. However due to its
111 huge number of features integration, it is not lightweight.