README.md

The third series of tests aim at evaluating the performance of the new
detector wrt the previous one on a selection of more standard images.
Compared measures $M$ are the execution time $T$, the amount $N$ of detected
blurred segments, the amount $N'$ of long (larger than 40 pixels) segments,
the mean length $L$ and the mean width $W$ of the detected segments.
For the sake of objectivity, these results are also compared to the same
measurements made on the image data base used for the CannyLine line
segment detector \cite{LuAl15}.
\RefTab{tab:auto} gives the measures obtained on one of the selected images
(\RefFig{fig:auto}) and the result of a systematic test on the whole
CannyLine data base.
\begin{figure}[h]
%\center
  \begin{tabular}{
      c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}}
    \includegraphics[width=0.32\textwidth]{Fig_auto/buro.png} &
    \includegraphics[width=0.32\textwidth]{Fig_auto/autoOld.png} &
    \includegraphics[width=0.32\textwidth]{Fig_auto/autoNew.png} \\
    & \includegraphics[width=0.22\textwidth]{Fig_auto/dssDetailOld.png} &
    \includegraphics[width=0.22\textwidth]{Fig_auto/dssDetailNew.png}
    \begin{picture}(1,1)
      {\color{red}{
        \put(-21.5,38){\framebox(18,8)}
        \put(-12.5,38){\vector(-2,-1){25}}
        \put(-135.5,38){\framebox(18,8)}
        \put(-124.5,38){\vector(-2,-1){25}}
      }}
      {\color{dwhite}{
        \put(-287,37.5){\circle*{8}}
        \put(-172,37.5){\circle*{8}}
        \put(-59,37.5){\circle*{8}}
        \put(-186,5.5){\circle*{8}}
        \put(-73,5.5){\circle*{8}}
      }}
      \put(-289.5,35){a}
      \put(-174.5,35){b}
      \put(-61.5,35){c}
      \put(-189,3){d}
      \put(-75.5,3){e}
    \end{picture}
  \end{tabular}
  \caption{Automatic detection on standard images:
           an input image (a), the segments found by the old detector (b)
           and those found by the new detector (c), and a detail of the
           enclosing digital segments for both old (d) and new (e) detectors.}
  \label{fig:auto}
\end{figure}
\begin{table}
\centering
\input{Fig_auto/perfTable}
\caption{Measured performance of both detectors on standard images.
         $M_{old}$ (resp. $M_{new}$) denotes the measure obtained with
         the previous (resp. new) detector.}
\label{tab:auto}
\end{table}

The new detector is faster and provides more blurred segments than the
previous one.
The details of \RefFig{fig:auto} d) and e) illustrate the improved
accuracy obtained.
The output segments are thinner but also shorter.
The control of the assigned width to fit to the detected segment width
has the side effect of blocking the segment expansion when the remote parts
are more noisy.
The relevance of this behavior depends strongly on application requirements.
Therefore the control of the assigned width is left as an option the user
can let or cancel.
In both case, it could be interesting to combine the detector with a tool
to merge aligned segments.

%Although these observations in unsupervised context should be reproduced
%in supervised context, similar experiments require an application context
%to dispose of a ground truth and of real users to assess the detector
%relevance through ergonomy evaluations.