Skip to content
Snippets Groups Projects
Commit e16c25f9 authored by even's avatar even
Browse files

Experiments first revisitation

parent a8b21264
No related branches found
No related tags found
No related merge requests found
\begin{tabular}{|l||r|r|r|r|r|}
\hline
Measure $M$ & \multicolumn{1}{c|}{$T$ (ms)} & \multicolumn{1}{c|}{$C$}
& \multicolumn{1}{c|}{$N$}
& \multicolumn{1}{c|}{$L$} & \multicolumn{1}{c|}{$L/N$} \\
\hline
Canny
& 75.4 $\pm$ 11.7 & 60.6 $\pm$ 10.6
& 466 $\pm$ 138 & 17678 $\pm$ 4419 & 39.5 $\pm$ 10.2 \\
Ours
& 83.2 $\pm$ 20.1 & 61.5 $\pm$ 10.8
& 613 $\pm$ 140 & 20769 $\pm$ 4000 & 34.6 $\pm$ 5.4 \\
\hline
\end{tabular}
This paper introduces a new straight line detector in gray-level images, This paper introduces a fully discrete framework for a new straight line
detector in gray-level images,
where line segments are enriched with a thickness parameter where line segments are enriched with a thickness parameter
intended to provide a quality criterion on the extracted feature. intended to provide a quality criterion on the extracted feature.
This study enhances previous works on interactive This study enhances previous works on interactive
line detection with a better estimation of the segment width and line detection with a better estimation of the segment width and
orientation through two main improvements: adaptive directional scans and orientation through two main improvements: adaptive directional scans and
the control of the assigned width to the detection algorithm. the control of the assigned width to the detection algorithm.
A new contribution to the detection of all the segments in a single image A new contribution to the automatic detection of all the segments in a single
is also proposed and left available in an online demonstration. image is also proposed and left available in an online demonstration.
...@@ -187,3 +187,28 @@ in urban imagery}, ...@@ -187,3 +187,28 @@ in urban imagery},
year = {2013}, year = {2013},
pages = {1578--1583} pages = {1578--1583}
} }
@article{HuertasMedioni86,
title = {Detection of intensity changes with subpixel accuracy using
{L}aplacian--{G}aussian masks},
author = {Huertas, Andres and Medioni, Gerard},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
volume = {8},
number = {5},
month = {September},
year = {1986},
pages = {651--664}
}
@article{KekreGharge10,
title = {Image segmentation using extended edge operator for mammographic
images},
author = {Kekre, H.B. and Gharge, S.M.},
journal = {International Journal on Computer Science and Engineering},
volume = {2},
number = {4},
year = {2010},
pages = {1086--1091}
}
...@@ -20,8 +20,10 @@ adaptive directional scans that continuously adjust the scan strip ...@@ -20,8 +20,10 @@ adaptive directional scans that continuously adjust the scan strip
to the detected edge direction; to the detected edge direction;
the control of the assigned width based on the observation of the the control of the assigned width based on the observation of the
blurred segment growth. blurred segment growth.
Expected gains in accuracy and execution time were confirmed by Experiments on synthetic ground truth images show improved accuracy
held experiments. brought be the new concepts, and a reliable extraction of the line width.
Moreover reached performance are quite comparable to a recent well
established edge detector in the literature.
A residual weakness of the approach is the sensitivity to the initial A residual weakness of the approach is the sensitivity to the initial
conditions. conditions.
......
\section{Experimental validation}
\label{sec:expe}
The main goal of this work is to detect straight segments enriched with a
quality measure through the associated width parameter.
In lack of available reference tool, the evaluation stage first aims
at quantifying the benefits of the new detector compared to the previous
one in unsupervised context on synthetic data considered as a ground truth.
Then comparisons are made with a well established recent detector
\cite{LuAl15} in order to check that global performance (processing time,
ground truth covering, detected lines count and mean length) are not
degraded.
%The process flow of the former method (initial detection followed by two
%refinement steps) is integrated as an option into the code of the new
%detector, so that both methods rely on the same optimized basic routines.
%During all these experiments, only the blurred segment size and its
%orientation compared to the initial stroke are tested at the end of
%the initial detection, and only the segment size is tested at the end
%of the fine tracking stage.
%All other tests, sparsity or fragmentation, are disabled.
%The segment minimal size is set to 5 pixels, except where precised.
The first test compares the performance of both
detectors on a set of 1000 synthesized images containing 10 randomly
placed input segments with random width between 2 and 5 pixels.
The absolute value of the difference of each found segment to its
matched input segment is measured.
On these ground-truth images, the numerical error on the gradient extraction
biases the line width measures. This bias was first estimated using 1000
images containing only one input segment (no possible interaction)
and the found value (1.4 pixel) was taken into account in the test.
\RefTab{tab:synth} shows
slightly better width and angle measurements for the new detector.
The new detector shows more precise, with a smaller amount of false
detections and succeeds in finding most of the input segments.
Other experiments, also available at the {\it GitHub} repository, show
that the new detector is faster and finds more edges than the previous one.
%\begin{figure}[h]
%\center
% \begin{tabular}{
% c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}c}
% \includegraphics[width=0.19\textwidth]{Fig_synth/statsExample.png} &
% \includegraphics[width=0.19\textwidth]{Fig_synth/statsoldPoints.png} &
% \includegraphics[width=0.19\textwidth]{Fig_synth/statsoldBounds.png} &
% \includegraphics[width=0.19\textwidth]{Fig_synth/statsnewPoints.png} &
% \includegraphics[width=0.19\textwidth]{Fig_synth/statsnewBounds.png}
% \begin{picture}(1,1)
% \put(-310,0){a)}
% \put(-240,0){b)}
% \put(-170,0){c)}
% \put(-100,0){d)}
% \put(-30,0){e)}
% \end{picture}
% \end{tabular}
% \caption{Evaluation on synthesized images:
% a) one of the test images,
% b) output blurred segments from the old detector and
% c) their enclosing digital segments,
% d) output blurred segments from the new detector and
% e) their enclosing digital segments.}
% \label{fig:synth}
%\end{figure}
\begin{table}
\centering
\input{Fig_synth/statsTable}
\caption{Measured performance of both detectors on a set of synthesized images.
$S$ is the set of all the input segments,
$D$ the set of all the detected blurred segments.}
\label{tab:synth}
\end{table}
The next experiments aim at evaluating the performance of the new
detector with respect to CannyLines detector \cite{LuAl15}
on the 102 annotated ground truth images of the York Urban data base
\cite{DenisAl08}.
%One of them is displayed on \RefFig{fig:auto}.
Compared measures $M$ are execution time $T$, covering ratio $C$,
Detected lines amount $N$, total lines length $L$ and ratio $L/N$.
To compare the execution times, for each image of the data base we measure
the execution time of 100 detections, gradient extraction included.
The covering ratio compares the length of ground truth lines covered
by the detected lines wrt the total length of ground truth lines.
A eight neighborhood of 8 pixels is added to lines provided by CannyLines.
\RefTab{tab:canny} gives the achieved results.
\begin{table}
\centering
\input{Expe_auto/cannyTable}
\caption{Measured performance of the proposed detector and CannyLines
on standard images.}
\label{tab:canny}
\end{table}
The new detector shows equivalent performance in terms of ground trouth lines
coverture. CannyLines is faster and finds less but longer lines, but these
results remain quite comparble so that we can arg that the global performance
is not degraded. And most of all, our detector provides an indication on the
detected lines quality through the additional thickness parameter.
...@@ -12,7 +12,7 @@ orientation. They are also used to detect structured features to help a ...@@ -12,7 +12,7 @@ orientation. They are also used to detect structured features to help a
Therefore, straight line detection is always an active research topic Therefore, straight line detection is always an active research topic
centered on the quest of still faster, more accurate or more robust-to-noise centered on the quest of still faster, more accurate or more robust-to-noise
methods \cite{AkinlarTopal12,GioiAl10,LuAl15,MatasAl00}. methods \cite{AkinlarTopal12,GioiAl10,LuAl15,MatasAl00}.
Most of the times, they are based on the extraction of an edge map based Most of the times, they rely on the extraction of an edge map based
on gradient magnitude. Gradient orientation is often used to discriminate on gradient magnitude. Gradient orientation is often used to discriminate
candidates and thus provide better efficiency. candidates and thus provide better efficiency.
However, they seldom provide an exploitable measure of the output line However, they seldom provide an exploitable measure of the output line
...@@ -29,10 +29,10 @@ complementary measures to reprojection errors for local accuracy evaluation. ...@@ -29,10 +29,10 @@ complementary measures to reprojection errors for local accuracy evaluation.
In digital geometry, new mathematical definitions of classical In digital geometry, new mathematical definitions of classical
geometric objects, such as lines or circles, have been developed geometric objects, such as lines or circles, have been developed
to better fit to the discrete nature of most of todays data to process. to better fit to the discrete nature of most of today's data to process.
In particular, the notion of blurred segment \cite{Buzer07,DebledAl05} was In particular, the notion of blurred segment \cite{Buzer07,DebledAl05} was
introduced to cope with the image noise or other sources of imperfections introduced to cope with the image noise or other sources of imperfections
from the real world by the mean of a width parameter. from the real world using a width parameter.
Efficient algorithms have already been designed to recognize Efficient algorithms have already been designed to recognize
these digital objects in binary images \cite{DebledAl06}. these digital objects in binary images \cite{DebledAl06}.
Blurred segments seem well suited to reflect the required line quality Blurred segments seem well suited to reflect the required line quality
......
...@@ -57,7 +57,7 @@ ...@@ -57,7 +57,7 @@
\input{methodV2} \input{methodV2}
\input{expe} \input{expeV2}
\input{conclusion} \input{conclusion}
......
...@@ -11,8 +11,8 @@ stable appearance such as metallic tubular objects \cite{AubryAl17}. ...@@ -11,8 +11,8 @@ stable appearance such as metallic tubular objects \cite{AubryAl17}.
Contrarily to most detectors, no edge map is built here, but gradient Contrarily to most detectors, no edge map is built here, but gradient
magnitude and orientation are examined in privileged directions to track magnitude and orientation are examined in privileged directions to track
edge traces. edge traces.
Therefore we use a Sobel operator with a 5x5 pixels mask \cite{GuptaMazumdar13} Therefore we use a Sobel operator with a 5x5 pixels mask
to get a high quality gradient information. \cite{KekreGharge10} to get a high quality gradient information.
\subsection{Previous work} \subsection{Previous work}
...@@ -42,8 +42,10 @@ The search direction relies on the support vector of the blurred segment ...@@ -42,8 +42,10 @@ The search direction relies on the support vector of the blurred segment
detected at the former step. detected at the former step.
Because the numerization rounding fixes a limit on this estimated orientation Because the numerization rounding fixes a limit on this estimated orientation
accuracy, more steps are inevitably necessary to process larger images. accuracy, more steps are inevitably necessary to process larger images.
In the following, we present the improvements in the new detector to
overcome these limitations.
\subsection{Workflow of the detection process} \subsection{Workflow of the new detection process}
The workflow of the detection process is summerized in the following figure. The workflow of the detection process is summerized in the following figure.
...@@ -68,11 +70,11 @@ In the fine tracking step, another blurred segment $\mathcal{B}'$ is built ...@@ -68,11 +70,11 @@ In the fine tracking step, another blurred segment $\mathcal{B}'$ is built
and extended with points that correspond to local maxima of the and extended with points that correspond to local maxima of the
image gradient, ranked by magnitude order, and with gradient direction image gradient, ranked by magnitude order, and with gradient direction
close to start point gradient direction. close to start point gradient direction.
At this refinement step, a control of the assigned width is applied At this refinement step, a {\it control of the assigned width} is applied
and an adaptive directional scan based on the found position $C$ and and an {\it adaptive directional scan} based on the found position $C$ and
direction $\vec{D}$ is used in order to extends the segment in the direction $\vec{D}$ is used in order to extends the segment in the
appropriate direction. These two improvements are described in the appropriate direction. These two improvements are described in the
following sections. following sections (\ref{subsec:ads} and \ref{subsec:caw}).
The output segment $\mathcal{B}'$ is finally tested according to the The output segment $\mathcal{B}'$ is finally tested according to the
application needs. Too short, too sparse or too fragmented segments application needs. Too short, too sparse or too fragmented segments
...@@ -82,6 +84,7 @@ intuitive parameters left at the end user disposal. ...@@ -82,6 +84,7 @@ intuitive parameters left at the end user disposal.
%to put forward achievable performance. %to put forward achievable performance.
\subsection{Adaptive directional scan} \subsection{Adaptive directional scan}
\label{subsec:ads}
The blurred segment is searched within a directional scan with a position The blurred segment is searched within a directional scan with a position
and an orientation approximately provided by the user, or blindly defined and an orientation approximately provided by the user, or blindly defined
...@@ -169,6 +172,7 @@ In practice, it is started after $\lambda = 20$ iterations when the observed ...@@ -169,6 +172,7 @@ In practice, it is started after $\lambda = 20$ iterations when the observed
direction becomes more stable. direction becomes more stable.
\subsection{Control of the assigned width} \subsection{Control of the assigned width}
\label{subsec:caw}
The assigned width $\varepsilon$ to the blurred segment recognition algorithm The assigned width $\varepsilon$ to the blurred segment recognition algorithm
is initially set to a large value $\varepsilon_0$ in order to allow the is initially set to a large value $\varepsilon_0$ in order to allow the
......
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