³æ¤¸¤»¡BÃä½t°»´ú

³¯¼yÃv

2004-11-03

±N¼v¹³¤Á³Î¦¨ª«¥ó©M­I´º¬Oª«¥ó¿ëÃÑ©M¼v¹³²z¸Ñ«eªº°ò¥»¤u§@¡A³o­Ó¤u§@¥D­nªº¥ô°È¬O§ä¥Xª«¥óªºÃä½t©M½u±ø¡C¹ï¤HÃþµøı¨t²Î(human visual system)¦Ó¨¥¡A³o¬O¤@¥ó»´¦Ó©öÁ|ªº¨Æ¡A¦ý¦b¹q¸£µøı¤¤¡AÁaµM¬O¬Û·í½ÆÂøªº²z½×©Mºtºâªk¡A³£¤£®e©ö¼Ò¥é¤HÃþªº³o­Ó¦æ¬°¯à¤O¡C¥D­nªº§xÃø¨Ó¦Û¥H¤U°ÝÃD¡G

¢w ­ì©l¼v¹³¶q¤Æªº»~®t©M·P´úªºÂø°T¡C¦¹¤@¦]¯À¾É­P°»´ú«G«×ÅܤƦý¨Ã«Dª«¥óÃä½tªº°Ï°ì¡A¦P®É¥i¯àº|¥¢¤@¨Çª«¥óÃä½t¦ý¬O¨S¦³©úÅã«G«×Åܤƪº°Ï°ì¡F

¢w ª«¥óÃä½tªº¥¿½T¦ì¸m·|¨ü¨ì¶q¤Æ»~®t©MÂø°Tªº¼vÅT¦Ó²£¥Í°¾²¾¡F

¢w ª«¥óÃä½t¦b¼v¹³¤¤À³ªí²{¥X¾U§Qªº«G«×ÅܤơA³o´N¬O¥¦ªº°ªÀW¯S©Ê¡A©Ò¥H¥ô¦ó·Q­n­°§C¼v¹³Âø°Tªº¥­·ÆÂoªi³£·|¾É­PÃä½t°Ï°ì°T¸¹ªº½k¤Æ(bluring)¡F

µ´¤j¦h¼ÆªºÃä½t°»´ú¤èªk³£¬O±Ä¥Î·L¤Àªº­ì²z¡A¦]¬°·L¤À·|©ñ¤j°ªÀW°T¸¹¡A¦ý¤]·|©ñ¤jÂø°T¡A¦]¦¹¥­·ÆÂoªiÁ`¬O»Ý­nªº¡C¥­·ÆÂoªiªºµ{«×¬O®Ú¾ÚÂoªi¾¹ªº¤j¤p(size)©MÂoªiªº¤Ø«×(scale)¡A¶V¤jªº¤Ø«×¶V¯àªí²{¥X¤j½d³òªº«G«×ÅܤơA¦ý¤]¨Ï±o©Ò°»´ú¥X¨ÓªºÃä½t¦ì¸mºë½T«×¶V®t¡F¸û¤pªº¤Ø«×Áö¯à§ä¥X¸û¬°¥¿½Tªº¦ì¸m¡A¦ý¤S®e©ö²£¦h³\¦h¿ù»~ªºÃä½tÂI¡C¤£¦P¼v¹³ªº¤Ø«×¼vÅT¤]¤£¬Û¦P¡A·Q­n§ä¥X¤@­Ó³Ì¨Îªº¤Ø«×¾A¥Î¦b¤@±i¼v¹³ªº¥þ³¡Ãä½t¬O«Ü§xÃøªº¡A·Q­n§ä¥X¤@­Ó³Ì¨Îªº¤Ø«×¾A¥Î¦b©Ò¦³¼v¹³§ó¬O¤£¥i¯àªº¨Æ¡C

¦h¤Ø«×ªºÃä½t°»´ú(multiscale edge detection)¤èªk¦]¦¹´£¨Ñ¤@­Ó¨ì¥Ø«e¬°¤î³Ì¦nªºÃä½t°»´ú¤èªk¡C³oºØ¤èªkªº®Ö¤ß·§©À¬OÀ³¥Î¤£¦P¤j¤pªº¥­·ÆÂoªi¾¹(¨Ò¦p°ª´µÂoªi¾¹)»P­ì©l¼v¹³§@°j±Û¿n¡A±o¨ì¤£¦PscaleªºÂoªi¼v¹³¡A¦A¹ï¨C¤@­ÓscaleªºÂoªi¼v¹³©â¨ú¨äÃä½t¡A³Ì«áÅ|¦X©Ò¦³scaleªºÃä½t¸ê°T¬°³Ì²×ªºÃä½t¼v¹³¡C

 

1. Scale spaceÂoªi

¤Ø«×ªÅ¶¡Âoªi¬O±´°Q°T¸¹(¼v¹³)ÀH¤Ø«×ÅܤƪºÂoªi§Þ³N¡C

­Y¦³¤@ºû°T¸¹f(x)¥H1-D°ª´µÂoªi¾¹¶i¦æ°j±Û¿nÂoªi¡G

ÅܤÆs¡A°T¸¹f(x)¦b(x, s)¥­­±¤W§Î¦¨ªº¹Ï§ÎºÙ¬°scale--space image(¤Ø«×ªÅ¶¡¼v¹³) F(x, s)¡G

F(x, s) = G(x, s)* f(x)

¹ï©ó¯S©ws0¡A¦±½uF(x, s0)ªº¤Ï¦±ÂI(inflection point)±Nº¡¨¬

 

±N©Ò¦³¤Ï¦±ÂI³s½u¥i§@¬°¤@²Õ(x, s)®y¼Ð¤Wªº¦±½u¡A¦p¤U¹Ï¡Cs¶V¤pªº¤è¦Vªí¥Üscale¸û¤jªº¤Ï¦±ÂI¡C

°Ñ¦Ò·j´M¡GIEEE XploreÃöÁä¦r¡Gscale space <and> edge

2. Ãä½t(Edge)ªº¯S©Ê

a. ¶¥ª¬Ãä½t­å­±¡Fb. ¥­·Æ¶¥ª¬Ãä½t­å­±+white noise(a signal with a flat frequency spectrum); c.¤@¶¥·L¤À­å­±¡Fd.¤G¶¥·L¤À­å­±¡F

a. ¼Ó±èª¬Ãä½t­å­±¡Fb.¤@¶¥·L¤À­å­±§e²{¨â­Ómaximum©M¤@­Óminimum¡Fd.¤G¶¥·L¤À­å­±§e²{¤T­Ózero-crossing¡F

3. CannyÃä½t°»´ú

Canny, J.F., A computational approach

to edge detection. IEEE Trans Pattern Analysis and Machine Intelligence, 8(6): 679-698, Nov 1986.

 

Canny´£¥XEdge DetectionÂoªi¾¹ªºµû¦ô·Ç«h¡G

1. ¦b¥­©Zªº¼v¹³°Ï°ì¨S¦³ÅTÀ³ => Âoªi¾¹«Y¼Æ©M¬°¹s:

2. µ¥¦V©Ê(Isotropy): Âoªi¾¹ÅTÀ³¥²¶·»Pedge¤è¦VµLÃö¡C

3. ¥¿½Tªºedge°»´ú¯à¤O: ÁקK¥H¤U±¡§Î

¢w±NÂø°T»~§P¬°edge¡A¦¹¬°false positive

¢w¥¼¯à§ä¥X¯u¥¿edge¡A¦¹¬°false negative

4. ¦nªº©w¦ì(localization): °»´úªºedge¦ì¸mÀ³¾¨¥i¯à±µªñ¯u¥¿edge¦ì¸m

5. ³æ¤@ÅTÀ³(single response): ¾¨¥i¯à´î¤Öedgeªþªñlocal maximaªº¼Æ¶q¡C

CannyªºÃä½t°»´ú¾¹­ì²z¡G

1. À³¥Î°ª´µÂoªi¾¹¦b¦Ç¶¥¼v¹³f(x,y)¡A±o¨ì¥­·Æ¼v¹³g(x; y) = f(x; y) * wG(x; y; £m)

2. À³¥Î·L¤ÀÂoªi¾¹¡¾g(x; y)­pºâÃä½t±j«×(magnitude)©M¤è¦V(orientation).

The scale parameter £m is selected based on

°ª´µÂoªi¾¹°Ñ¼Æ£m¨M©wÃä½t°»´ú¾¹ªº¤j¤p¡C¨M©w¾A·í£mÀ³¨Ì¤U¦C»Ý¨D¡G

¢w ·Q­n±o¨ìedgeªº²Ó¸`µ{«×(fine edges vs global edges);

¢w Âø°Tªº¦h¹è;

¢w °»´ú©Ê©w¦ì/·Ç½T«×trade off

Canny´£¥X¨â­Ó¤èªk¨Óº¡¨¬Ãä½t°»´ú¾¹³æ¤@ÅTÀ³©M©w¦ì¥¿½T©Êªº­n¨D¡G

¢w Non-maxima suppression

1. ¹ï¨C¤@ÂIC(x; y), ¿ï©w««ª½©óorientation¤è¦V¨â­Ó°¼Ã䪺¾FªñÂI¡A°O§@A©MB¡F.

2. ¦pªGM(A) > M(C) or M(B) > M(C), «hC¤£¬°edge(³]©wM(C(x,y))=0);

3. ¿é¥X(edge)±j«×¼v¹³MNMS(x; y)

non-maxima suppression¿é¥X¤´·|±a¦³¤@¨Ç«Dedgeªºlocal maxima¡A¦P®É connectivity©Ê½è¤£©úÅ㪺edge°Ï°ì¡CCannyªºHysteresis thresholding¤èªk´£¨Ñ¤F¸Ñ¨M¤è®×¡C

¢w Hysteresis thresholding

1. ©w¸q¨â­Óthresholds, Thigh and Tlow¡A

¹³¯À(x; y) ¦pªGMNMS (x; y) > Thigh¡A¸Ó¹³¯À´NºÙ¬°strong¡A

¹³¯À(x; y) ¦pªGMNMS (x; y)¡ØTlow¡A¸Ó¹³¯À´NºÙ¬°weak¡A

©Ò¦³¨ä¥Lªº¹³¯ÀºÙ¬°candidate¡F

2. ¦pªG¹³¯À(x; y)¬Oweak¡A«h²¤¥h¡F¦pªG¬Ostrong¡A«h¿é¥X¬°edge¹³¯À¡F

3. ¦pªG¹³¯À(x; y)¬Ocandidate¡A¦Ó¥BMNMS > Tlow¡A«h§PÂ_¬O§_ªuµÛlocal maxima¬Û³sªºedge¤è¦V¦³¬ï¹L(x,y)¡A­Y¬O¡A«h¿é¥X¬°edge¡F

4. ¦pªGcandidate¹³¯À(x; y)»P¤@strong¹³¯À¡A«h¿é¥X¸Ócandidate¬°edge¡C


  • hysteresis thresholdingªº¶i¤@¨B»¡©ú¡G
    • If any edge response is above a high threshold, those pixels constitute definite output of the edge detector for a particular scale.
    • Individual weak responses usually correspond to noise.
    • Such connected pixels are treated as edge pixels if their response is above a low threshold.
    • The low and high thresholds are set according to an estimated signal to noise ratio.

CannyÃä½t°»´ú¾¹»Pscale space

n         The correct scale for the operator depends on the objects contained in the image.

n         The solution to this unknown is to use multiple scales and aggregate information from them.

n         Different scale for the Canny detector is represented by different standard deviations£mof the Gaussians.

n         There may be several scales of operators that give significant responses to edges (i.e., signal to noise ratio above the threshold); in this case the operator with the smallest scale is chosen as it gives the best localization of the edge.


CannyÃä½t°»´úÂoªiºtºâªk¡G

  1. Åܤƣm­«½Æ¨BÆJ(2)¨ì(6)¡F
  2. ¥Hscale £mªº°ª´µÂoªi¾¹¹ïimage g°õ¦æ°j±Û¿n¡F

3.     ¦ô´ú¨C¤@¹³¯Àªº°Ï°ìÃä½tªº¥¿¥æ¤è¦Vn

4.     À³¥Înon-maximal suppression¤èªk§ä¥Xedges¦ì¸m¡F

  1. ­pºâedges±j«×MNMS (x; y)
  2. À³¥Îhysteresis thresholding®ø°£edge¤òÃä¡A¨Ã±µÄòedgeÂ_ÂI¡F
  3. ²Ö¿n¦h­«scale £mªºedge¸ê°T¡A¦X¦¨³Ì«áªºedge¼v¹³¡C

 

Ãä½t°»´ú§@¬°¤@­Ó³]­p³Ì¨Î¤Æªº°ÝÃD¡A¨äµû¦ô¨ç¼Æ¬°¡G

1. Maximize the signal to noise ratio to give good detection. This favours the marking of true positives.

2. Achieve good localization to accurately mark edges.

3. Minimize the number of responses to a single edge. This favours the identification of true negatives, that is, non-edges are not marked.

 

CannyÃä½t°»´ú¾¹µ{¦¡½X¤U¸ü

 

¥»¶g²ßÃD¡G

1. ¨Ï¥ÎCannyÃä½t°»´ú¾¹¡A¿ï©w¤@²Õ¾A·íªº Thigh and Tlow¡A¦A¤À§O¥Hs=0.5, 1.0, 2.0, 3.0¡A¨D±oant(gray)600x400¼v¹³4­Óscaleªºedge¼v¹³¡A±N¨ä¦X¦¨¬°¤@±i³Ì²×ªºedge¼v¹³(¤£¤@©w¨Ï¥Î­«Å|¬Û¥[(OR)ªº¤è¦¡)¡C

2. ¨ú¥X­ì©l¼v¹³ªº¨ä¤¤¤@±ø¤ô¥­profile¡A¨Ï¥Î¬Û¦Pµ{¦¡¡A¨D±o¦p²Ä¤@¸`©Ò³¯­zªº¤Ø«×ªÅ¶¡¼v¹³¡C

½u°»´ú(Line Detection)°ÝÃD

½u«¬ºA(line pattern)¤£¦P©óÃä½t«¬ºA(edge pattern)ªº¯S¼x¡A¦b¼v¹³¤¤¡A½u«¬ºA¸g±`ªí²{¦b¤@¨Ç¼e«×«Ü¯¶ªºvalley/ridge¯¾¸ô¼Ë¦¡©Î¬O¤â¼g¤å¦r¡Bø¹Ï¼v¹³¤¤¡C¦pªG¥¦¨ã¦³¬YºØ¼e«×¡A«hÂǥѼзÇÃä½t°»´ú¤èªk±`·|§ä¥X¨â±ø¥­¦æªº½u±ø¡A³o¬O¦]¬°¤j³¡¤ÀªºÃä½t°»´ú¤èªk³£«Ø¥ß¦b¶¥ª¬¨ç¼Æ(step edge)ªº°ò¦¡C¤£¦P©óedgeÁ`¬O¦ì©ó¤G¶¥·L¤Àªºzero-crossing¦ì¸m¡Aline«h©ó¤@¶¥·L¤Àªºzero-crossingªº¦ì¸m¡C

Steger´£¥X¨Ï¥ÎGaussian©M¤@¦¸·L¤ÀªºÂoªi¤èªk§@¬°½u°»´ú¾¹(cf. C. Steger : An unbiased Detector of Curvilinear Structure, IEEE Transaction on PAMI, 20(2), 1998); Ziou(Djemel Ziou : Optimal Line detection, 2000)«h±Ä¥ÎCannyªºÃä½t°»´úµû¦ô·Ç«h¡A´£¥X¥HIIRÂoªi¾¹¨Ó°»´ú½u¯S¼x¡C

3. ½Ð¨Ï¥Î²Ä¤­³æ¤¸ªºRobinson¤@¶¥¾É¼ÆÂoªi¾¹¨D¥Xfinger300x300«ü¯¾¼v¹³ªºamplitude©Morientation¼v¹³¡A(A)§Q¥Îamplitude¼v¹³§ä¥Xzero-crossingªº¹³¯À«á¿é¥Xzero-crossing¼v¹³(¶Â¥Õ¼v¹³)¡C(B)§Q¥Îorientation¼v¹³¸ê°T¡Aµe¥X¨C¤@­Ózero-crossing¦ì¸mªº¤è¦V¹Ï¦p¤U¹Ï(b)¡F

         

(a)                          (b)

(C)À³¥ÎCannyªºnon-maximal suppression©Mhysteresis thresholding¤èªk§@¬°½u°»´ú«á³B²z¡A¿é¥X³Ì«áªºline image¡C