Article: remarks BK updated

Article/Expe_auto/perfTable.tex

 \begin{tabular}{|l||r|r|r|r|}
 \hline
-Measure $M$ & \multicolumn{1}{c|}{$T$ (ms)} & \multicolumn{1}{c|}{$N$}
-& \multicolumn{1}{c|}{$L$ (pixels)} & \multicolumn{1}{c|}{$W$ (pixels)} \\
+Measure $M$ & \multicolumn{1}{c|}{$T$} & \multicolumn{1}{c|}{$N$}
+& \multicolumn{1}{c|}{$L$} & \multicolumn{1}{c|}{$W$} \\
 \hline
-$M_{old}$ on image of \RefFig{fig:auto} & 29.51 & 306 & 38.58 & 2.47 \\
-$M_{new}$ on image of \RefFig{fig:auto} & 25.85 & 352 & 33.25 & 2.17 \\
-$M_{new}/M_{old}$ (\%)
+$M_{old}$ on image of \RefFig{fig:auto}
+& 29.51 ms & 306 segments & 38.58 pixels & 2.47 pixels \\
+$M_{new}$ on image of \RefFig{fig:auto}
+& 25.85 ms & 352 segments & 33.25 pixels & 2.17 pixels \\
+\hline \hline
+$M_{new}/M_{old}$ (\%) & & & & \\
+\hspace{0.4cm} on image of \RefFig{fig:auto}
 & \multicolumn{1}{l|}{87.60} & \multicolumn{1}{l|}{115.03}
 & \multicolumn{1}{l|}{86.18} & \multicolumn{1}{l|}{87.85} \\
-\hline
-$M_{new}/M_{old}$ (\%) & & & & \\
-on CannyLines images
+\hspace{0.4cm} on the set of test images & & & & \\
+\hspace{0.4cm} on CannyLines images
 & 86.02 $\pm$ 2.44 & 110.15 $\pm$ 6.51 & 89.23 $\pm$ 5.11 & 84.70 $\pm$ 2.98 \\
 \hline
 \end{tabular}

Article/Expe_hard/readme.txt

@@ -7,3 +7,7 @@ m : detection auto avec le nouveau detecteur
 p : sauve l'image capture.png -> ismm/Expe_hard/hardNew.png
 6 : detection auto avec l'ancien detecteur
 p : sauve l'image capture.png -> ismm/Expe_hard/hardOld.png
+
+Pour les images hardDetailNew.png et hardDetailOld.png,
+selectionner le coin haut gauche de l'image depuis le
+point (346, 140).

Article/Fig_notions/bswidth.tex

 \begin{picture}(220,60)
   \multiput(0,6)(2,-6){2}{\line(3,1){150}}
-  \multiput(0,8)(30,6){8}{\color{red}{\line(5,1){10}}}
-  \multiput(0,18)(30,6){8}{\color{red}{\line(5,1){10}}}
+  \multiput(0,8)(30,6){8}{\color{blue}{\line(5,1){10}}}
+  \multiput(0,18)(30,6){8}{\color{blue}{\line(5,1){10}}}
   \put(45,21){\circle*{3}}
   \put(55,21){\circle*{3}}
   \put(65,21){\circle*{3}}
@@ -15,8 +15,8 @@
   \put(140,36){\vector(0,1){10}}
   \put(140,62.66){\vector(0,-1){10}}
   \put(120,58){$\mu_{i-1}$}
-  \put(160,30){\color{red}{\vector(0,1){10}}}
-  \put(160,60){\color{red}{\vector(0,-1){10}}}
-  \put(164,36){\color{red}{$\mu_i$}}
-  \put(180,60){\color{red}{$\mathcal{B}_{i}$}}
+  \put(160,30){\color{blue}{\vector(0,1){10}}}
+  \put(160,60){\color{blue}{\vector(0,-1){10}}}
+  \put(164,36){\color{blue}{$\mu_i$}}
+  \put(180,60){\color{blue}{$\mathcal{B}_{i}$}}
 \end{picture}

Article/Fig_notions/readme.txt

+Seg_adaption est le code qui a fourni adaption*_*.png
+a partir de test_adapt.txt
+0 pour lancer le test
+1 pour ouvrir la vue de structure
++ pour zoomer une fois
+se caler en bas a droite avec les fleches
+i - i pour voir les scans bounds
+p pour le screenshot -> structure.png
+I pour voir les scans
+p pour le screenshot -> structure.png
+Avec gimp, selection de la zone (516,70) (174,250)
+
+
+
+Le meme exemple a servi a produire une premiere figure escape
+sur le segment P1 (181, 226), P2 (178, 205)
+Avec gimp, selection de la zone (159,115) (355,203)
+
+
+
+Seg_escape est le code qui a fourni escape*_*.png
+a partir de test_escape.txt
+En vue principale :
+Ctrl-q pour basculer entre scan statique ou adaptatif
+En vue d'analyse (touche 1) :
+i pour selectionner les vues de scan
+Ctrl-j pour visualiser la selection utilisateur
+Avec gimp, selection de la zone (333,227)(344,208)
+
+
+Production des images plus claires :
+6 y pour avoir un niveau de sombreur a 30
+Avec gimp, selection de la zone (58,127)(520,275)
+pour avoir des images 454x148 a l'arrivee.

Article/Fig_synth/res.txt

-1000 cartes avec 1000 segments noirs sur fond blanc
-re-estimation de la largeur en retirant un biais de 1.4
-
   RESULTS FOR THE OLD DETECTOR
-69.719 (pm 20.3805) segments searches (local min) / image
-25.23 (pm 6.65638) provided segments / image
-10.766 (pm 1.96395) provided long segments / image
-2.298 (pm 2.82935) undetected segments per image
-88.3997 (pm 4.31352) % of points found
-2.27666 (pm 1.50916) % of points found more than once (redetections)
-32.2193 (pm 13.8132) % false points produced
-Precision : 0.732884 (pm 0.0844506)
-Biased width : 4.57922 (0.38789) per matched segment
-Width : 3.37696 (0.350989) per matched segment
-Width difference : 0.64275 (0.319964) per matched segment
-Absolute width difference : 0.874094 (0.288227) per matched segment
-Angle difference : 0.0497484 (0.756378) degrees per matched segment
-Absolute angle difference : 1.2933 (0.92141) per matched segment
-Absolute long edge angle difference : 0.44657 (0.55775) per matched segment
+66.841 (pm 23.0353) segments searches (local min) / image
+25.351 (pm 7.16695) provided segments / image
+10.41 (pm 1.84318) provided long segments / image
+2.527 (pm 2.54081) undetected segments per image
+Recall : 89.0911 (pm 3.94178) % of points found
+2.40181 (pm 1.52098) % of points found more than once (redetections)
+34.8326 (pm 16.4621) % false points produced
+Precision : 0.718919 (pm 0.0973014)
+Statistical precision : 0.727566 (pm 0.0969161)
+Statistical recall : 0.891995 (pm 0.0394028)
+Statistical F-measure : 0.798543 (pm 0.0677593)
+Biased width : 4.64042 (0.340364) per matched segment
+Width : 3.42451 (0.334641) per matched segment
+Width difference : 0.717033 (0.34932) per matched segment
+Absolute width difference : 0.920479 (0.309156) per matched segment
+Angle difference : -0.246099 (1.33358) degrees per matched segment
+Absolute angle difference : 1.48162 (1.41907) per matched segment
+Absolute long edge angle difference : 0.542687 (1.08617) per matched segment
   RESULTS FOR THE NEW DETECTOR
-63.6 (pm 17.4179) segments searches (local min) / image
-23.722 (pm 4.90028) provided segments / image
-11.619 (pm 2.08619) provided long segments / image
-0.475 (pm 0.756272) undetected segments per image
-89.813 (pm 3.17276) % of points found
-1.83739 (pm 1.3875) % of points found more than once (redetections)
-22.5376 (pm 10.2339) % false points produced
-Precision : 0.799399 (pm 0.0701648)
-Biased width : 4.49693 (0.320961) per matched segment
-Width : 3.26288 (0.297711) per matched segment
-Width difference : 0.408348 (0.27149) per matched segment
-Absolute width difference : 0.741648 (0.24851) per matched segment
-Angle difference : 0.0354951 (0.641921) degrees per matched segment
-Absolute angle difference : 0.982936 (0.696344) per matched segment
-Long edge absolute angle difference : 0.480284 (0.568648) per matched segment
-
+62.393 (pm 17.1718) segments searches (local min) / image
+24.351 (pm 5.02624) provided segments / image
+11.145 (pm 1.91824) provided long segments / image
+0.637 (pm 0.909978) undetected segments per image
+Recall : 90.0194 (pm 2.77084) % of points found
+1.78012 (pm 1.21784) % of points found more than once (redetections)
+23.7471 (pm 8.87023) % false points produced
+Precision : 0.791264 (pm 0.0630463)
+Statistical precision : 0.791896 (pm 0.0630431)
+Statistical recall : 0.900763 (pm 0.0277026)
+Statistical F-measure : 0.84169 (pm 0.0416617)
+Biased width : 4.5098 (0.252717) per matched segment
+Width : 3.28158 (0.239712) per matched segment
+Width difference : 0.462984 (0.255638) per matched segment
+Absolute width difference : 0.764102 (0.231734) per matched segment
+Angle difference : -0.141146 (0.721153) degrees per matched segment
+Absolute angle difference : 1.05426 (0.801184) per matched segment
+Long edge absolute angle difference : 0.529121 (0.684013) per matched segment

Article/Fig_synth/statsTable.tex

@@ -3,18 +3,20 @@
 Detector : & \multicolumn{3}{c|}{old} & \multicolumn{3}{c|}{new} \\
 \hline
 Detected blurred segments per image
-& 25.23 & $\pm$ & 6.66 & 23.72 & $\pm$ & 4.90 \\
+& 25.35 & $\pm$ & 7.17 & 24.35 & $\pm$ & 5.03 \\
 Detected long (> 40 pixels) blurred segments per image
-& 10.77 & $\pm$ & 1.96 & 11.62 & $\pm$ & 2.09 \\
+& 10.41 & $\pm$ & 1.84 & 11.14 & $\pm$ & 1.92 \\
 Undetected input segments per image
-& 2.30 & $\pm$ & 2.83 & 0.47 & $\pm$ & 0.76 \\
-Ratio of true detection (\%) : $\#(D\cap S)/\#S$
-& 88.40 & $\pm$ & 4.31 & 89.81 & $\pm$ & 3.17 \\
-Precision (\%) : $\#(D\cap S)/\#D$
-& 73.29 & $\pm$ & 8.45 & 79.94 & $\pm$ & 7.02 \\
+& 2.53 & $\pm$ & 2.54 & 0.64 & $\pm$ & 0.91 \\
+Precision (\%) : $P = \#(D\cap S)/\#D$
+& 72.76 & $\pm$ & 9.69 & 79.19 & $\pm$ & 6.30 \\
+Recall (ratio of true detection) (\%) : $R = \#(D\cap S)/\#S$
+& 89.20 & $\pm$ & 3.94 & 90.08 & $\pm$ & 2.77 \\
+F-measure (harmonic mean) (\%) : $F = 2\times P\times R/(P+R)$
+& 79.85 & $\pm$ & 6.78 & 84.17 & $\pm$ & 4.17 \\
 Width difference (in pixels) to matched input segment
-& 0.87 & $\pm$ & 0.29 & 0.74 & $\pm$ & 0.24 \\
+& 0.92 & $\pm$ & 0.31 & 0.76 & $\pm$ & 0.23 \\
 Angle difference (in degrees) to matched input segment
-& 1.29 & $\pm$ & 0.92 & 0.98 & $\pm$ & 0.70 \\
+& 1.48 & $\pm$ & 1.42 & 1.05 & $\pm$ & 0.80 \\
 \hline
 \end{tabular}

Article/biblio.bib

@@ -32,7 +32,7 @@
 
 
 @inproceedings{DebledAl05,
-  title = {Blurred Segments Decomposition in Linear Time},
+  title = {Blurred segments decomposition in linear time},
   author = {Debled-Rennesson, Isabelle and Feschet, Fabien and
             Rouyer-Degli, Jocelyne},
   booktitle = {Proc. of Int. Conf. on DGCI},
@@ -43,7 +43,7 @@
   pages = {371-382},
   optaddress = {Poitiers, France},
   optmonth = {April},
-  publisher = {Springer}
+  optpublisher = {Springer}
 }
 
 

Article/conclusion.tex

@@ -3,36 +3,39 @@
 \label{sec:conclusion}
 
 This paper introduced a new straight edge detector based on a local analysis of
-the image gradient and on the use of blurred segments to vehiculate an
+the image gradient and on the use of blurred segments to embed an
 estimation of the edge thickness.
 It relies on directional scans of the image around maximal values of the
 gradient magnitude, that have previously been presented in
 \cite{KerautretEven09}.
-Despite of good performances achieved, the former approach suffers of two
-major drawbacks: the inaccurate estimation of the blurred segment width
-and orientation, and the lack of guarantee that it is completely detected.
-These limitations were solved through the integration of two new concepts:
+%Despite of good performances achieved, the former approach suffers of two
+%major drawbacks: the inaccurate estimation of the blurred segment width
+%and orientation, and the lack of guarantee that it is completely detected.
+%These limitations were solved through the integration of two new concepts:
+%adaptive directional scans that continuously adjust the scan strip
+%to the detected blurred segment direction;
+The main limitations of the former approach were solved through the integration
+of two new concepts:
 adaptive directional scans that continuously adjust the scan strip
-to the detected blurred segment direction;
+to the detected edge direction;
 the control of the assigned width based on the observation of the
-blurred segment thickenning in the early stage of its expansion.
+blurred segment growth.
+Expected gains in accuracy and execution time were confirmed by the
+held experiments.
 
-Expected gains in execution time linked to the suppression of a useless
-repetition of the fine tracking stage were confirmed by the experiments
-both in supervised and unsupervised contexts.
-The residual weakness is the high sensitivity to the initial conditions
-despite of the valuable enhancement brought by the duplication of the
-initial detection.
-Disturbing gradient perturbations in the early stage of the edge expansion,
-possibly due to the presence of close edges, can deeply affect the output
-blurred segment.
-In supervised context, the user can easily select a favourable area where
+A residual weakness of the approach is the sensitivity to the initial
+conditions.
+In supervised context, the user can select a favourable area where
 the awaited edge is dominant.
-But this default remains quite sensible in unsupervised context.
-
-In future works, we intend to provide some solutions for this drawback
-by scoring the detection result on the base of a characterization of the
-initial context.
+This task is made quite easier, thanks to the stabilization produced by
+the duplication of the initial detection.
+But in unsupervised context, gradient perturbations in the early stage of
+the edge expansion, msotly due to the presence of close edges, can deeply
+affect the result.
+In future works, we intend to provide solutions to this drawback
+by scoring the detection result on the basis of a characterization of the
+local context.
+%
 Then experimental validation of the consistency of the estimated width and
 orientation values on real situations are planned in different application
 fields.

Article/expe.tex

@@ -11,6 +11,12 @@ For a fair comparison, the process flow of the former method (the 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 were tested at the end of
+the initial detection, and only the segment size was tested at the end
+of the fine tracking stage.
+All the other tests, sparsity or fragmentation, were disabled.
+The segment minimal size was set to 5 pixels, except where precised.
 
 The first test (\RefFig{fig:synth}) compares the performance of both
 detectors on a set of 1000 synthesized images containing 10 randomly
@@ -18,20 +24,14 @@ 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 such perfect image, the numerical error on the gradient extraction
-biases the line width measures. This bias was estimated in quasi ideal
-context (only one input segment, thus no risk of perturbation)
+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.
-Although the effect of the old detector weaknesses should be reduced in
-this low gradient noise context, the results of \RefTab{tab:synth} show
+The results of \RefTab{tab:synth} show
 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.
 
-The second test (\RefFig{fig:hard}) visually compares the results of both
-detectors on quite noisy images, also difficult to process for other
-detectors from the literature. 
-The new detector provides less outliers and misaligned segments, and
-globally more relevant informations to infere the structure of the brick wall.
 \begin{figure}[h]
 %\center
   \begin{tabular}{
@@ -65,29 +65,7 @@ $S$ is the set of all the input segments,
 $D$ the set of all the detected blurred segments.}
 \label{tab:synth}
 \end{table}
-\begin{figure}
-%\center
-  \begin{tabular}{
-      c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}}
-    \includegraphics[width=0.32\textwidth]{Fig_method/parpaings.png} &
-    \includegraphics[width=0.32\textwidth]{Expe_hard/hardOld.png} &
-    \includegraphics[width=0.32\textwidth]{Expe_hard/hardNew.png}
-    \begin{picture}(1,1)
-      {\color{dwhite}{
-        \put(-286,4.5){\circle*{8}}
-        \put(-171,4.5){\circle*{8}}
-        \put(-58,4.5){\circle*{8}}
-      }}
-      \put(-288.5,2){a}
-      \put(-173.5,2){b}
-      \put(-60.5,2){c}
-    \end{picture}
-  \end{tabular}
-  \caption{Detection results on quite textured images:
-           one of the tested images (a),
-           the segments found by the old detector (b)
-           and those found by the new detector (c).}
-  \label{fig:hard}
-\end{figure}
 
 \input{expeAuto}
+
+\input{expeHard}

Article/expeAuto.tex

-
-The third campaign of tests aims at evaluating the performance of the new
-detector with respect to the previous one on a selection of more standard
-images.
+The next experiments aim at evaluating the performance of the new
+detector with respect to the previous one on a test set composed of a
+selection of 20 real images.
+One of them is displayed on \RefFig{fig:auto}.
 Compared measures $M$ are the execution time $T$, the amount $N$ of detected
-blurred segments, the mean length $L$ and the mean width $W$ of the detected
-segments.
+blurred segments, their mean length $L$ and their mean width $W$.
 For the sake of objectivity, these results are also compared to the same
 measurements made on the 20 images 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 CannyLine
-data base.
+\RefTab{tab:auto} gives the achieved results.
 \begin{figure}[h]
 %\center
   \begin{tabular}{
@@ -20,7 +17,8 @@ data base.
 %    \includegraphics[width=0.32\textwidth]{Expe_auto/coloredNew.png} \\
     \includegraphics[width=0.32\textwidth]{Expe_auto/bsOld.png} &
     \includegraphics[width=0.32\textwidth]{Expe_auto/bsNew.png} \\
-    & \includegraphics[width=0.22\textwidth]{Expe_auto/dssDetailOld.png} &
+    \includegraphics[width=0.22\textwidth]{Expe_auto/buroDetail.png} &
+    \includegraphics[width=0.22\textwidth]{Expe_auto/dssDetailOld.png} &
     \includegraphics[width=0.22\textwidth]{Expe_auto/dssDetailNew.png}
     \begin{picture}(1,1)
       {\color{red}{
@@ -28,25 +26,30 @@ data base.
         \put(-5.5,31){\vector(-2,-1){20}}
         \put(-133.5,31){\framebox(28,9)}
         \put(-117.5,31){\vector(-2,-1){20}}
+        \put(-247.5,31){\framebox(28,9)}
+        \put(-231.5,31){\vector(-2,-1){20}}
       }}
       {\color{dwhite}{
         \put(-291,32.5){\circle*{8}}
         \put(-177,32.5){\circle*{8}}
         \put(-63,32.5){\circle*{8}}
+        \put(-302,4.5){\circle*{8}}
         \put(-188,4.5){\circle*{8}}
         \put(-75,4.5){\circle*{8}}
       }}
       \put(-293.5,30){a}
       \put(-179.5,30){b}
       \put(-65.5,30){c}
-      \put(-191,2){d}
-      \put(-77.5,2){e}
+      \put(-305,2){d}
+      \put(-191,2){e}
+      \put(-77.5,2){f}
     \end{picture}
   \end{tabular}
-  \caption{Automatic detection on standard images:
+  \caption{Automatic detection on real 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.}
+           image (d) and the enclosing digital segments for both old (e)
+           and new (f) detectors.}
   \label{fig:auto}
 \end{figure}
 \begin{table}
@@ -69,7 +72,7 @@ Found edges are thus more fragmented.
 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 it.
-In both case, it could be interesting to combine the detector with a tool
+In both case, it could be useful to combine the detector with a tool
 to merge aligned segments.
 
 %Although these observations in unsupervised context should be reproduced

Article/expeHard.tex

-
-The second test (\RefFig{fig:hard}) visually compares the results of
-both detectors on quite difficult images with a lot of gradient noise. 
-The new detector provides less outliers and misaligned segments, and
-globally more relevant informations to infere the structure of the brick wall.
-\begin{figure}[h]
-%\center
+The last test visually compares the results of both detectors on quite textured
+images, also difficult to process for other detectors from the literature. 
+The minimal size parameter was raised to 12 pixels to reject small segments
+considered as outliers.
+On the example of \RefFig{fig:hard}, the new detector provides less residual
+outliers and misaligned segments, and globally more relevant informations
+to infere the structure of the brick wall.
+\begin{figure}
+  \center
   \begin{tabular}{
       c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}c@{\hspace{0.1cm}}}
-    \includegraphics[width=0.32\textwidth]{Fig_method/parpaings.png} &
-    \includegraphics[width=0.32\textwidth]{Fig_hard/hardOld.png} &
-    \includegraphics[width=0.32\textwidth]{Fig_hard/hardNew.png}
+    \includegraphics[width=0.205\textwidth]{Fig_method/parpaings.png} &
+    \includegraphics[width=0.32\textwidth]{Expe_hard/hardDetailOld.png} &
+    \includegraphics[width=0.32\textwidth]{Expe_hard/hardDetailNew.png}
     \begin{picture}(1,1)
+      {\color{red}{
+        \put(-302,34){\framebox(29,11.5)}
+      }}
       {\color{dwhite}{
-        \put(-286,4.5){\circle*{8}}
+        \put(-266,4.5){\circle*{8}}
         \put(-171,4.5){\circle*{8}}
         \put(-58,4.5){\circle*{8}}
       }}
-      \put(-288.5,2){a}
+      \put(-268.5,2){a}
       \put(-173.5,2){b}
       \put(-60.5,2){c}
     \end{picture}
   \end{tabular}
-  \caption{Detection results on quite textured images:
-           one of the tested images (a),
-           the segments found by the old detector (b)
-           and those found by the new detector (c).}
+  \caption{Results on quite textured images: test image (a),
+           detail (top left corner) on the segments found by the old
+           detector (b) and on those found by the new detector (c).}
   \label{fig:hard}
 \end{figure}