2 \subsection{Image Quality}
3 The visual quality of the STABYLO scheme is evaluated in this section.
4 Four metrics are computed in these experiments :
5 the Peak Signal to Noise Ratio (PSNR),
6 the PSNR-HVS-M familly~\cite{PSECAL07,psnrhvsm11} ,
7 the BIQI~\cite{MB10,biqi11} and
8 the weigthed PSNR (wPSNR)~\cite{DBLP:conf/ih/PereiraVMMP01}.
9 The first one is widely used but does not take into
10 account Human Visual System (HVS).
11 The other last ones have been designed to tackle this problem.
15 \begin{tabular}{|c|c|c|}
17 Embedding rate & Adaptive
22 PSNR-HVS-M & 78.6 & 72.9 \\
26 wPSNR & 86.43& 77.47 \\
30 \caption{Quality measeures of our steganography approach\label{table:quality}}
34 Compare to the Edge Adpative scheme detailed in~\cite{Luo:2010:EAI:1824719.1824720}, our both wPSNR and PSNR values are always higher than their ones.
36 \JFC{comparer aux autres approaches}
40 \subsection{Steganalysis}
44 The quality of our approach has been evaluated through the two
45 AUMP~\cite{Fillatre:2012:ASL:2333143.2333587}
46 and Ensemble Classifier~\cite{DBLP:journals/tifs/KodovskyFH12} based steganalysers.
47 Both aims at detecting hidden bits in grayscale natural images and are
48 considered as the state of the art of steganalysers in spatial domain~\cite{FK12}.
49 The former approach is based on a simplified parametric model of natural images.
50 Parameters are firstly estimated and a adaptive Asymptotically Uniformly Most Powerful
51 (AUMP) test is designed (theoretically and practically) to check whether
52 a natural image has stego content or not.
53 In the latter, the authors show that the
54 machine learning step, (which is often
55 implemented as support vector machine)
56 can be a favourably executed thanks to an Ensemble Classifiers.
62 \begin{tabular}{|c|c|c|c|}
63 Shemes & \multicolumn{2}{|c|}{STABYLO} & HUGO\\
65 Embedding rate & Adaptive & 10 \% & 10 \%\\
67 AUMP & 0.39 & 0.22 & 0.50 \\
69 Ensemble Classifier & & & \\
74 \caption{Steganalysing STABYLO\label{table:steganalyse}}