saguar1

YANAI Lab.

電気通信大学 総合情報学科/大学院 総合情報学専攻 メディア情報学コース 柳井研究室
電気通信大学 > 情報工学科 > コンピュータ学講座 > 柳井研究室 > 研究紹介  

$B2hA|Cf$N0aN`NN0h$+$i$N?'>pJsCj=P(B

$BAjED(B $BM%(B

2012$BG/(B 2$B7n(B 10$BF|(B




1 $BGX7J(B

$B6aG/!"l$K$h$C$F(BWeb$B>e$NpJs$r4^$`%G!<%?$rBgNL$K=8$a$k$3$H$,2DG=$G$"$k!#(B $B$^$?!"%G!<%?%^%$%K%s%0$K$h$kBgNL%G!<%?2r@O$,%^!<%1%F%#%s%0$NJ,Ln$G=EMW;k$5$l$F$*$j!"(B $B2hA|$N;k3P>pJs$rMxMQ$7$?%^%$%K%s%0pJs$rCj=P$9$k$3$H$,2DG=$H$J$l$P!"%G!<%?%^%$%K%s%0pJs$NCj=P$O=EMW$J;v9`$H$J$C$F$$$k!#(B

2 $BL\E*(B

$BK\8&5f$NL\E*$O!"2hA|Cf$K4^$^$l$k0aN`NN0h$+$i!"4{B8$N%G!<%?%^%$%K%s%0pJs$rCj=P$9$k$3$H$G$"$k!#(B $B;k3P>pJs$NCf$G$bFC$K?'$KCeL\$7!"2hA|Cf$N0aN`$N?'$r(B12$B?'$KJ,N`$9$k$r=PNO$9$k!#(B $B2hA|Cf$N0aN`$NNN0h$r?dDj$7!"?'J,N`$r9T$&$3$H$G0aN`$N?'L>$,%i%Y%k$H$7$F=PNO$5$l$k$,!":G=*E*$J%i%Y%k$N7hDj$O0aN`NN0h$NLL@Q%Y!<%9$K$h$k7hDj

3 $B4XO"8&5f(B

$BF02hCf$N?MJ*$,Ce$F$$$k%7%c%D$N?'J,N`$r9T$C$?$b$N$H$7$F(BWui$B$i$N8&5f$d(BVaquero$B$i$N8&5f$,5s$2$i$l$k(B[#!wui06!#][#!vaq09!#]$B!#(B $B$3$l$i$N8&5f$NL\E*$O!"KIHH!&4F;k%7%9%F%`$G$N1~MQ$d!"8D?M8!:w$r9T$&$H$$$&E@$K$"$j!"$=$N$?$a$K%7%c%D$N?'J,N`$r9T$C$F$$$k!#(B $BK\8&5f$GMQ$$$k%G!<%?$OF02h$G$O$J$/@E;_2h$H$J$k!#(B $B$^$?!"%7%c%D$@$1$G$J$/!"%:%\%s!"%V!<%D!"%P%C%0$H$$$C$?!"?MJ*$,CeMQ$7$F$$$kJ*BNA4$F$,?'J,N`$NBP>]$H$J$k!#(B

$B?'J,N`$r9T$C$?8&5f$H$7$F(BWeijer$B$i$N8&5f$,5s$2$i$l$k(B[#!colorNames!#]$B!#(B eBay$B%G!<%?%;%C%H$r;HMQ$7!"(BPLSA$B$d(BSVM$B$H$$$C$?pJs$rMxMQ$7$?J}$,@:EY$,$h$$$3$H$,J,$+$C$F$$$k!#(B $BK\.NN0hC10L$G$N>pJs$rMQ$$$k!#(B

4 $BDs0F

4.1 $B35MW(B

$B$$!"2hA|Cf$K4^$^$l$k0aN`$N?'J,N`$r9T$&!#(B
  1. $B2hA|$rF~NO$9$k(B
  2. $BA0=hM}$H$7$F2hA|$N?'Jd@5$d%/%j%C%T%s%0$r9T$&(B
  3. $B4i8!=P$r9T$$!"?MJ*$,
  4. $BNN0hJ,3d$r9T$$!"?MJ*$HGX7J$NJ,N%$r9T$&(B
  5. $BNN0h$r:YJ,2=$7!"0aN`NN0h$N%9!<%Q!<%T%/%;%k$rF@$k(B
  6. $B3F>.NN0hKh$K2hA|FCD'$rCj=P$7!"%i%Y%k$r$D$1$k(B
  7. $BF1$8%i%Y%k$r;}$D%9!<%Q!<%T%/%;%kF1;N$r7k9g$9$k(B
  8. $BBg$-$JNN0h$r$H$J$C$?%i%Y%k$rA*=P$7!"$=$l$r=PNO$9$k(B
$B1$B$K<($9!#(B
$B?^(B 1: $B4i8!=P!ANN0h:YJ,2=$N=hM}$NN.$l(B
\includegraphics[width=\hsize]{img/seg.eps}

4.1.0.1 $BA0=hM}(B

$B2hA|Cf$N0aN`$O8w$d1F$N1F6A$rl9g$b$3$N$H$-9T$&!#(B

4.1.0.2 $B4i8!=P(B

$BK\8&5f$G07$&0aN`$N2hA|$OCeMQ$5$l$F$$$k$b$N$H$7!"0aN`NN0h$N?dDj$O4i8!=P$+$i;O$^$k!#(B $B$3$3$G$O!"(BHaar-like$BFCD'$rMQ$$$?%V!<%9%F%#%s%0$K$h$kJ,N`4o$rMxMQ$7!"4i8!=P$r9T$C$F$$$k!#(B $B$^$?!"8e=R$9$kNN0hJ,3d

4.1.0.3 $B?MJ*!&GX7J$NJ,N%(B

$B$3$3$G$ONN0hJ,3d.2=$9$k$3$H$G!"J*BN$HGX7J$K2hA|$rJ,3d$9$k!#(B $B%(%M%k%.!<4X?t$O!"NY@\2hAG$N:9J,CM$H!"2hAG$,J*BN!&GX7J$KB0$9$k3NN(%b%G%k$r;H$C$FDj5A$5$l$k!#(B $BGX7J$KB0$9$k2hAG=89g$r;XDj$7!"$=$l$i$NCM$r%,%&%9:.9gJ,I[$K$h$C$FI=8=$9$k$3$H$G3NN(%b%G%k$,7hDj$5$l$k!#(B $BA0=R$7$?4i8!=P%9%F%C%W$GBg$^$+$JNN0h$r7hDj$9$kI,MW$,$"$k$N$O!"$3$N$?$a$G$"$k!#(B

4.1.0.4 $BNN0h:YJ,2=(B

$B0aN`C10L$G$N=hM}$r9T$&$?$a!"NN0h$r:YJ,2=$9$k!#:YJ,2=$5$l$?2hA|NN0h$O%9!<%Q!<%T%/%;%k$H8F$P$l$k!#(B $B%9!<%Q!<%T%/%;%k$N:n@.$K$O(BFelzenszwalb$B$i$,Ds0F$7$?%0%i%U%Y!<%9$NNN0hJ,3d.A40hLZ$r9=@.$7$F$$$/2aDx$G!"E,;~!"LZ$NJ,CG$r9T$&$3$H$GNN0hJ,3d$rC#@.$9$k

4.1.0.5 $B>.NN0hC10L$G$N%i%Y%kIU$1(B

$B3F%9!<%Q!<%T%/%;%k$+$i2hA|FCD'$rCj=P$7!"?'J,N`$r9T$&$3$H$G!"%i%Y%kIU$1$r9T$&!#(B $B$3$3$G!"%i%Y%k$OBeI=E*$J(B12$B?'$N?'L>$rI=$9$b$N$H$9$k!#(B $B$3$l$K$h$C$F!"3F%9!<%Q!<%T%/%;%k$,(B12$B?'$NFb$N$$$:$l$+$N?'L>$r;}$D$3$H$K$J$k!#(B

4.1.0.6 $B>.NN0h$NE}9g(B

$B$3$3$G$OF1$8%i%Y%k$r;}$D%9!<%Q!<%T%/%;%kF1;N$r7k9g$7$F$$$/!#(B $B$3$N=hM}$K$h$j!"J#?t8D$"$C$?%9!<%Q!<%T%/%;%k$,!":GBg$G$b(B12$B8D$^$G$K$^$H$a$i$l$k!#(B 12$B8D$NNN0h$O$=$l$>$l!"(B12$B?'$NFb$N$$$:$l$+$N?'$,@j$a$kNN0h$rI=$9$3$H$K$J$k!#(B $BNN0hE}9g$N7k2L!"?^(B2$B$N$h$&$J%^%9%/$,F@$i$l$k!#(B

$B?^(B 2: $BE}9g8e$N%9!<%Q!<%T%/%;%k(B
\includegraphics[width=\hsize]{img/joinResult/blue.eps} blue
\includegraphics[width=\hsize]{img/joinResult/white.eps} white
\includegraphics[width=\hsize]{img/joinResult/gray.eps} gray
\includegraphics[width=\hsize]{img/joinResult/brown.eps} brown

4.1.0.7 $B%i%Y%k$N=PNO(B

$B:G=*E*$J%i%Y%k$OLL@Q%Y!<%9$N7hDjK!$K$h$C$F5a$a$i$l$k!#(B $BBg$-$JNN0h$H$J$C$F$$$k$b$N$@$1$rA*$S!"$=$N%i%Y%k$r=PNO$9$k!#(B 12$B8D$NNN0h$NAmOB$rDj?tG\(B(0:1]$B$7!"$=$l$rogCM$H$7$F%i%Y%k$NA*Br$r9T$C$?!#(B $B$3$l$K$h$j!"?^(B2$B$NNc$+$i$O!"(Bgray$B!"(B brown$B$N$h$&$J%i%Y%k$,GK4~$5$l!"(Bblue$B!"(Bwhite$B$N%i%Y%k$N$_$,A*$P$l$k$3$H$,4|BT$G$-$k!#(B

4.2 $B?'J,N`(B

$B?'J,N`$O5!3#3X=,K!$K$h$j9T$&!#(B $B3F>.NN0hKh$K2hA|FCD'$rCj=P$7!"$=$l$i$rI=(B1$B$G<($9(B12$B?'$N$$$:$l$+$KJ,N`$9$k!#(B


$BI=(B 1: $BJ,N`$9$k(B12$B?'0lMw(B
yellow orange green gray purple pink
brown black blue beige white red

$B2hA|FCD'$H$7$F!"3F?'6u4V$K$*$1$k%+%i!<%R%9%H%0%i%`$H(BBag-of-Color(BoC)$B%R%9%H%0%i%`$rCj=P$7!"@:EYHf3S$r9T$C$?!#(B $BDL>o$N%+%i!<%R%9%H%0%i%`$G$O?'6u4V$,Ey4V3V$K6h@Z$i$lEjI<$,9T$o$l$k$3$H$K$J$k$,!"(B BoC$B$G$O!"?'$NJ,I[$rI=8=$9$k$?$a$N=EMWEY$K1~$8$F?'6u4V$,J,3d$5$l$k$3$H$K$J$k!#(B $B$^$?!"J,N`4o$H$7$F!"(BK-Nearest-Neighbor(KNN)$B$H(BSupport Vector Machine(SVM)$B$rMQ$$!"@:EYHf3S$r9T$C$?!#(B KNN$B$O%7%s%W%k$J%"%k%4%j%:%`$G$"$k$,!"%G!<%?9=B$$r9)IW$9$k$3$H$G9bB.$J8!:w$,2DG=$G$"$j!"(B $BBg5,LO%G!<%?$G$NMxMQ$K4|BT$,=8$^$C$F$$$k!#(B SVM$B$O?'J,N`$K4X$9$k4{B88&5f$G$bMQ$$$i$l$?

5 $B

5.1 $B%G!<%?%;%C%H(B

$BK\8&5f$G$O!"%U%!%C%7%g%s%5%$%H(B1$B$+$i<}=8$7$?(B9592$BKg$N?MJ*2hA|$rBP>]$H$7!"

5.2 $B%^%9%/$,M?$($i$l$?$H$-$N?'J,N`@:EY(B

$B:n@.$7$?%G!<%?%;%C%H$rMQ$$$F!"%^%9%/$,M?$($i$l$?$H$-$N?'J,N`@:EY$r5a$a$?!#(B $B%^%9%/$G;XDj$5$l$?NN0h$+$i2hA|FCD'$rCj=P$7!"(B12$B?'$N3X=,!&J,N`$r9T$$!"(B 5-fold$B$N%/%m%9%P%j%G!<%7%g%s$K$h$j(B12$B?'$NJ,N`N($r5a$a$?!#(B $B$3$l$O!"0aN`NN0h$,3Nr7o$K$*$$$F!"$I$NDxEY$N?'J,N`$,2DG=$G$"$k$+$r3N$+$a$k$?$a$N$l!"%R%9%H%0%i%`$O(B1000$B3$B$K<($9!#(B

$B%+%i!<%R%9%H%0%i%`$H(BSVM$B$K$*$$$F!"(BRGB$B?'6u4V$G$N7k2L$,:G$b$h$/!"$=$N$H$-$NJ,N`N($O(B65.7%$B$H$J$C$?!#(B BoC$B$H(BSVM$B$K$h$kJ,N`$K$*$$$F$b(BRGB$B?'6u4V$,:G$b$h$/!"$=$N$H$-$NJ,N`N($O$,(B63.5%$B$H$J$C$?!#(B KNN$B$rMQ$$$?J,N`$G$O!"(BK$B$NCM$rJQ2=$5$;

$B?^(B 3: $B%^%9%/$,M?$($i$l$?$H$-(B
\includegraphics[width=\hsize]{img/mask_result.eps}
$B?^(B 4: $BDs0F
$B?^(B 5: $BDs0F

5.3 $BDs0F

$BK\8&5f$NDs0F$r5a$a!"$=$l$i$NNN0h$rE}9g$7!"NN0h$NBg$-$5$r$b$H$K7hDj$5$l$k!#(B 12$B8D$NNN0h$NAmOB$KDj?tG\(B(0:1]$B$r$+$1$?$b$N$rogCM$H$7$FMQ$$!"$=$l0J2<$N$b$N$rGK4~$9$k$3$H$G%i%Y%k$N:NBr$,9T$o$l$k!#(B $B$=$N$H$-$NDj?tG\$NCM$r(Brate$B$H$7!"(Brate$B$NCM$rJQ$(!":F8=N((B(recall)$B$HE,9gN((B(precision)$B$r5a$a$?!#(B recall$B$H(Bprecision $B$O
\begin{displaymath}
recall = \frac{$B@5$7$/J,N`$5$l$?$b$N$NAm?t(B}{$B8!=P$5$l$k$Y$-%i%Y%k$NAm?t(B} \\
\end{displaymath} (1)


\begin{displaymath}
precision = \frac{$B@5$7$/J,N`$5$l$?$b$N$NAm?t(B}{$B8!=P$5$l$?%i%Y%k$NAm?t(B}
\end{displaymath} (2)

$B%9!<%Q!<%T%/%;%kKh$G$N?'J,N`$G$O!"%+%i!<%R%9%H%0%i%`$H(BSVM$B$K$h$k4$B!"(B5$B$K<($9!#(B rate$B$,(B0$B$N$H$-$O8!=P$7$??'$9$Y$F$r:NMQ$7$?$H$-$HF1Ey$G$"$j!"(Brate$B$,(B1$B$N$H$-$O!"E}9g$5$l$?%9!<%Q!<%T%/%;%k$,M#0l$N?'$K9J$i$l$k$H$-$H$J$k!#(B

F$BCM$,:G9b$H$J$C$?$N$O(BRGB$B$K$*$$$F(Brate=0.13$B$H$7$?$H$-$G$"$j!"$=$N$H$-$N:F8=N($O(B61.0%$B$G$"$j!"E,9gN($O(B56.7%$B$H$J$C$?!#(B $B$3$N$H$-$N%i%Y%k$N8!=P?t$O(B298$B7o$G$"$j!"$=$NFb@52r$H$J$C$?$b$N$O(B169$B7o$H$J$k!#(B $B$=$l$KBP$7!"%^%9%/$rM?$($?$H$-$Nr7o$K$*$$$F$b!"?'L>$K$h$k%i%Y%kIU$1$N@-G=Dc2<$,>/$J$$$3$H$,J,$+$k!#(B $B%^%9%/$rM?$($?$H$-$NJ,N`N($N:G9bCM$,(B65.7%$B$G$"$k0J>e!"K\

6 $B$*$o$j$K(B

$BK\8&5f$G$O!"2hA|Cf$N0aN`NN0h$+$i$N?'>pJsCj=P $BK\.$5$9$.$kNN0h$N:o=|$r9T$&$H$$$C$?2~NI0F$,$"$k!#(B $B1F$NNN0h$d!"0aN`$H0aN`$N6-L\$OHs>o$K>.$5$J%9!<%Q!<%T%/%;%k$H$J$k798~$,$"$k!#(B $B$=$&$7$?%9!<%Q!<%T%/%;%k$O!"0aN`$N?'$NH=Dj$K$*$$$F%N%$%:$H$J$k2DG=@-$,9b$/!"$3$l$i$r:o=|$9$k$3$H$G@:EY8~>e$,8+9~$a$k!#(B $B$^$?!"?'J,N`$O5!3#3X=,K!$K$h$kJ,N`$H$J$k$,!"$=$3$G3X=,$H$7$F;H$&%G!<%?$Ne$,8+9~$a$k!#(B

section*$B;29MJ88%;29MJ88%;29MJ88%(B 1-0.8mm 0.1pt enumiv 4000 4000 `.

1
G. Wui, A. Rahimi, K. Goh, T. Tsai, A. Jain, Y. Wu, E. Y. Chang, and Y. F. Wang.
Identifying color in motion in video sensors.
In Proc. of IEEE Computer Vision and Pattern Recognition, pp. 561-569, 2006.

2
D. A. Vaquero, R. S. Feris, D. Tran, L. Brown, A. Hampapur, and M. Turk.
Attribute-based people search in surveillance environments.
In Proc. of Workshop on Applications of Computer Vision, pp. 1-8, 2009.

3
J. Van De Weijer, C. Schmid, and J. Verbeek.
Learning color names from real-world images.
In Proc. of IEEE Computer Vision and Pattern Recognition, pp. 1-8, 2007.

4
C. Rother, V. Kolmogorov, and A. Blake.
Grabcut: Interactive foreground extraction using iterated graph cuts.
In Proc. of ACM SIGGRAPH, pp. 309-314, 2004.

5
P. F. Felzenszwalb and D. P. Huttenlocher.
Efficient graph-based image segmentation.
International Journal of Computer Vision, 2004. Empty `thebibliography' environment



$B5SCm(B

... $BK\8&5f$G$O!"%U%!%C%7%g%s%5%$%H(B1
http://eruca.jp/