Skip to content
Snippets Groups Projects
Commit f8133ed0 authored by even's avatar even
Browse files

Article: experiments

parent c5609bc7
No related branches found
No related tags found
No related merge requests found
Showing
with 135 additions and 22 deletions
Article/Fig_expe/autoNew.png

122 KiB

Article/Fig_expe/buro.png

639 KiB

Article/Fig_expe/buroNew.png

554 KiB

Article/Fig_expe/outliersNew_full.png

94.3 KiB

Article/Fig_expe/outliersNew_zoom.png

23.8 KiB

494 221
585 201
515 467
521 427
135 70
156 33
610 500
621 477
743 165
787 133
102 124
105 89
526 524
539 494
92 365
91 349
391 84
414 85
92 365
91 349
\section{Experiments}
Mettre en valeur les gains obtenus par rapport \`a la version pr\'ec\'edente
avec des comparatifs en temps d'ex\'ecution et en qualit\'e de d\'etection
avec notamment une meilleure estimation de l'orientation.
D\'efauts persistants :
\begin{itemize}
\item L'\'epaisseur trouv\'ee n'est pas certifi\'ee.
\item Le r\'esultat d\'epend des conditions initiales.
Ca reste une m\'ethode instable, m\^eme si la duplication de la premi\`ere
\'etape a permis de gagner en stabilit\'e.
\item On n'est pas \`a l'abri d'un contour voisin qui vient perturber
la d\'etection initiale ou l'affinement.
Les filtres en fin de tracking sont l\`a pour soigner, pas pour gu\'erir.
\end{itemize}
\section{Experimental validation}
The evaluation stage aims at quantifying the advantages of the new detector
compared to the former one.
For a fair comparison, the process flow of the former method (the initial
detection followed by two refinement steps) was coded as an option of the
new detector, because since that time, the code of basic routineshas
largely been improved.
For instance, the new directional scanners are encoded as an iterator so that
only required scan lines are provided when required, whereas with the former
code, all the scan lines were computed and returned, whenever they were used
or not.
We just checked that the outputs of the new coding of the former detector
are equivalent to those of the old release.
\begin{figure}[h]
\center
\begin{tabular}{c@{\hspace{0.2cm}}c}
\includegraphics[width=0.49\textwidth]{Fig_expe/buro.png} &
\includegraphics[width=0.49\textwidth]{Fig_expe/buroNew.png}
\begin{picture}(1,1)
\put(-18,30){\circle{8}}
\put(-22,26){\makebox(8,8){\scriptsize 1}}
\put(-57,92){\circle{8}}
\put(-61,88){\makebox(8,8){\scriptsize 2}}
\put(-53,104){\circle{8}}
\put(-57,100){\makebox(8,8){\scriptsize 3}}
\put(-89,49){\circle{8}}
\put(-93,45){\makebox(8,8){\scriptsize 4}}
\put(-92,23){\circle{8}}
\put(-96,19){\makebox(8,8){\scriptsize 5}}
\put(-134,9){\circle{8}}
\put(-138,5){\makebox(8,8){\scriptsize 6}}
\put(-156,27){\circle{8}}
\put(-160,23){\makebox(8,8){\scriptsize 7}}
\put(-157,82){\circle{8}}
\put(-161,78){\makebox(8,8){\scriptsize 8}}
\put(-39,110){\circle{8}}
\put(-43,106){\makebox(8,8){\scriptsize 9}}
\end{picture}
\end{tabular}
\caption{Outputs of both former (on left) and new (on right) detectors
using a selec tionof input strokes.}
\label{fig:buro}
\end{figure}
The first test compares the computation times of both detectors on a
selection of input strokes (\RefFig{fig:buro}). Results are displayed
in table \RefTab{tab:cmpOldNew}.
\begin{table}
\centering
\begin{tabular}{|l||l|l|l|l|l|l|l|l|l|l|}
\hline \multicolumn{1}{|r||}{Stroke \hspace{0.4cm}} &
\multicolumn{1}{c|}{1} & \multicolumn{1}{c|}{2} &
\multicolumn{1}{c|}{3} & \multicolumn{1}{c|}{4} &
\multicolumn{1}{c|}{5} & \multicolumn{1}{c|}{6} &
\multicolumn{1}{c|}{7} & \multicolumn{1}{c|}{8} &
\multicolumn{1}{c|}{9} & \multicolumn{1}{c|}{10} \\ \hline \hline
with the former detector: \hspace{0.4cm}
& 18.2 & 18.2 & 18.2 & 18.2 & 18.2 & 18.2 & 18.2 & 18.2 & 18.2 & 18.2 \\ \hline
with the new detector: & & & & & & & & & & \\ \hline
\end{tabular}
\caption{Compared execution time in milliseconds between former and new
detectors with the input strokes of \RefFig{fig:buro}.}
\label{tab:cmpOldNew}
\end{table}
In the second series of tests, we compare the execution times of both detectors
for the automatic detection of edges on a set of test images. We display the
results for one of them (\RefFig{fig:auto}). X (resp. Y) blurred segments are
extracted with the former (resp. new) detector on all images. The
average execution time is X ms for the former detector, and Y ms for the
new detector.
\begin{figure}[h]
\center
\begin{tabular}{c@{\hspace{0.2cm}}c}
\includegraphics[width=0.49\textwidth]{Fig_expe/autoNew.png} &
\includegraphics[width=0.49\textwidth]{Fig_expe/autoNew.png}
\end{tabular}
\caption{Automatic edge detections on one of the test images with the
former detector on the left, and the new detector on the right.}
\label{fig:auto}
\end{figure}
The former detector do not estimate the edge width, but just circumscribes
the edge with a blurred segment of assigned width.
If the edge is very thin, the blurred segment is free to rotate around the
extracted edge and the provided orientation is biased.
Moreover it lets some space to incorporate additional spurious outliers.
With the new appoach, a real estimation of the edge width is provided.
The main risk of outlier incorporation remains at the beginning of the
blurred segment expansion as long as the minimal width continues to grow
and the assigned width has not been set to the detected segment minimal width.
\begin{figure}[h]
\center
\begin{tabular}{c@{\hspace{0.2cm}}c}
\includegraphics[width=0.49\textwidth]{Fig_expe/outliersNew_zoom.png} &
\includegraphics[width=0.49\textwidth]{Fig_expe/outliersNew_zoom.png}
\end{tabular}
\caption{Potential insertion of outliers for both detectors:
On the left, the fixed width of the former detector always lets opportunities
of outlier insertions. On the right, the new detector restricts these
opportunities to the blurred segment early analysis stage.}
\label{fig:outliers}
\end{figure}
\newcommand{\RefFigure}[1]{Fig.\,\ref{#1}}
\newcommand{\RefSection}[1]{Section\,\ref{#1}}
\newcommand{\RefFig}[1]{Fig.\,\ref{#1}}
\newcommand{\RefSec}[1]{Section\,\ref{#1}}
\newcommand{\RefTab}[1]{Tab.\,\ref{#1}}
......@@ -129,7 +129,7 @@ edge points are assumed to be oriented in the same direction. But if the
sign of the gradient direction is not considered, points with gradient in
opposite directions are merged to build the same blurred segment, allowing
the detection of both edges of a thin linear structure, like for instance
the tile joins of \RefFigure{fig:edgeDir}.
the tile joins of \RefFig{fig:edgeDir}.
\begin{figure}[h]
\center
......@@ -147,13 +147,13 @@ the tile joins of \RefFigure{fig:edgeDir}.
\end{figure}
On that example, when a straight features detection is run
(\RefFigure{fig:edgeDir} a)),
(\RefFig{fig:edgeDir} a)),
a thick blurred segment which extends up to four tiles is provided.
When a straight edge detection is run, a very thin blurred segment is
built to follow only one join edge.
The multi-detection can also be applied to both thin object or edge detection.
In the latter case, the detection algorithm is run twice using opposite
directions, so that in the exemple of figure (\RefFigure{fig:edgeDir} b)),
directions, so that in the exemple of figure (\RefFig{fig:edgeDir} b)),
both edges (in different colours) are highlighted.
These two thin blurred segments are much shorter, probably because the
tiles are not perfectly aligned.
......@@ -170,7 +170,7 @@ to collect all the segments found under the stroke.
\input{Fig_method/algoAuto}
The behaviour of the unsupervised detection is depicted through the two
examples of \RefFigure{fig:auto}.
examples of \RefFig{fig:auto}.
The example on the left shows the detection of thin straight objects on a
circle with variable width.
On the left half of the circumference, the distance between both edges
......
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment