In figure 3.8 the estimated location £ (eq. 2) are plotted for all
pixels. The underlying multiscale image is that of figure 3.1. At a 49 re. a MO nie ie eee
......... A OO enneKssnonnae TT
o= 0 the position of the pixels without point estimation are 40. 0.00.00 "esses Neneees sus eseKeuvt ense
marked for reference. It is interesting to note that the order of MAIS SEE ctttensskenuous to tavta kesataeno
4,)-"o0vaesous n acasocoKeceous.ns veeukesseseen e. vesusce
»* c*ecessoKeeeoonv v scouckecsoooccs
the estimated location £ was never changed, i.e. for all pixels i 4f. t
secoscsco 0 veuconeKounose 0 cocceKecosoceeo .......
the relation x(i) = x(i+1) produces £(i) s £(i+1). The equal sign 45. Si ei tt dir
: ; A OO o. cocovcoc
was only observed at o= 0. The dislocation of edges over scale 44. sese tuueyfenenusia «e rud uut tend us peus .
is easy to see at the dominant edge in figure 3.8. 43 Rr e iz en
........ A OO os cscciesssnen ee ...Z.Z.ZZ
4.2 -
GO 230- EA EA ema 1 000 0 N gcesssseseke sessio veseeKsssevens o. eacus"
gi- teme m tenens Xeecseco 8 vecesfesesssse ® ceooccoc
22.0 - .........>...>.>.. Mosssuce soc. n ca»ccsocsoosecescocce ° : wseseoee o coccococco)ecooee © 3 ovcecKoeveoseoo e cocenec
4.0- 0.0.0... e e cecccoceKecceso o cococecoscocn o 0000000
210 - A © 00000000000000008 BEER Inareofontwena e Tu avsven
3.9-
20.0- -* sesssenecescsosceXsesceconseo n ceevecsnnsenacos aswessse 0 000 seen Men sue: 0 D anvanNoen anus een...
3.0- €vs09500 v» cooncess)fesceoo © .....Ä.....>.... esveece
] 190- e esvssssrrsszavssisnasivesssusis caesoesteetonse pocrenes Cu RS à VE eme v etats eandra
3.7 -
190" =. v vsesesevescoseocKeenesesesen n vecdesovkosues sefisssne e seminis sosikensen n. vo iesesKensonsosio, 0 200008
36= soffencoe anne a nen Kennun u ae vsenMfenceuecs © .......
N @ sesasesssenss 13 DL ILLI ——————
IDS nave svpvepessnvibus ian Keeseesean eden exe e diurine MM Lm
d4- .okeseo «anon naeciuenpo mob loserisssevese oJ eteevon
ini- 450 - eros s o eure vesesesecoKsceceonese sie avoue ...... 33. = 8450 © vie n 20.0 Weser > mendmencironnenseretanpesnpe
e is 140- vouceus » sevscenscececcuKrecenveusson 0 queccenscue ie naines © ro e coccieoso + cen c odKesoosos Pes vuvnnes
t in $312] 2 unurrenn u muasdesesereneKeessasesess 8 aste enne eee t Wt COO en
31- seXKecoo © diee c» ceceKecoo c de c cd eoooooeooc ...>... .
3.0- eokeoss = Wee » erue S Mb u ed Renee .......
une rr ....0 » 0... ..... 0.0 10.0 20.0 30.0 40.0 50.0
'ery
12.0 - 999099009 © v9920e0000000)Xv02902200092009 9 59909992900
10.0- avococonoe c vesoccoceceKeceoconceses .... .......... pixel
9.0- 000000000 » ee e» ecco» Yoeononoaene ..... © sesssse ..
8.0- ono »eccosococ)itrsovosescecvoeososo eo nsossace
7.0- ...... ° eo. ——— 000900000000 © 00000000 OG 2.0- LLL NNNM ML ME DIE
D IMME © un a vôlfre » spe © dffesoano sesses sees
af © of o ceo cdjeeo © cde o diécseoeo nee 00n0 c
5.0- 00000... A TO » ceccoce 1.6 - dies © ab o oo odéco o cde © Jeveoe ann 0000000 oe
6.0 ~ 000020000 © cnccccofenscsunssessscseasncs c sssesss 1.8 -
40- 0.0.0.0... A A ° 0... 1.4 - dies © cd o c doo c» eo diesen ocsseemsmneme c
3.0- see» c dés © coo meo o Po © c dKe enero 0000000 4.2 - Hee © coco cde © memo conooosoccoccce
2.0- je © db o co © eder ede + dfondocces esse cc 1.0 - des ni wie mw eU 2. uec uvoceccnosuae ow
4.0 — rr ———— coe 0.8 -
dés + |o —X- Jm a. mmféwes * conne suc 0000000
ÜD- conoo0ossou0 .......>............. ................ 0.6 4
do 10.0 29.0 39.0 400 S0. 0.0 10.0 20.0 30.0 40.0 50.0
^ pe — pixel
2- Figure 3.8 Location of edge candidates in scale space. The de-
tected edges are marked by * Figure 3.9 Detailed view on the estimated locations £ in scale
space (0 range is 0.7-5.0, a02= 0.5)
The splitting process of an edge (at x = 30) with increasing
; Again, figure 3.9 gives a detailed view of the locations x esti-
ung 8 E E scale can be observed at the level o2 1.4.
in mated with subpixel position. The tracking of the edges from
be coarse to fine seems not a severe problem with these small
the differences between the scale levels. Between o= 2 and o= 3
we left a gap. The missing information caused by this gap does
not rise any reasoning problems in this example.
4 TRACKING AND MATCHING IN REAL IMAGES
e is
The task of tracking images through scale space and the conse-
ves
quences for matching will be demonstrated with examples of
real images. The real images are taken from Hahn and Fórstner
(1988). The stereo pair is given in epipolar geometry. The real
data example used consists of corresponding rows from these
images. The scale space images are visualized in figures 4.11
and 4.1r. The appended 1 and r denote the left and right stereo
partners, respectively.
).5)
299