Andrew Bibitchev
e subpixel accuracy;
e absence of vague thresholds.
Also, three is serious disadvantage — rather high calculation cost. To eliminate this drawback Gauss pyramid can be
used (see Section 3).
2.3 Fusion Of Integral Intensity Step And Correlation
Let step? (as a’ ) be integral step transform of the “source” image j? (c5 5): step” (a P a”) be integral step
transform of the “destination” image i” (x? D ); la? o? ) N be selected local maximums of step” (a S );
las bi 1% be conjectural end points of 2d edge L(a? Jaf ). To obtain set of 3d edges {4B} one can apply the
following procedure:
1. For each i=1..N andeach j=1..K,; we perform maximization
E443 8. uon S 45 SY, oo Ol1D 38 Sen LAS of aN s $. qw
W, (d ,Q :) = [step (a ,0 )+ step (a (d'A 0),0 (da o) corr[Pr,s s s, a5 Prs gs os, b 10) —> max
di S 35 sS sS 8 S
with restrictions d -di|s2Ad',|a" -aj|s2Aa , 0eQ. (14)
Here Pr, p is projection of point p on line L. As a result we obtain as a; ; 0; and a; = Pr,s Za) a ;
e NAT 74
TS. s
2. Foreachi=1..N wechoose ; with maximum value of @;.a; 0) . Let me denote such j as 72 ;
After all, we can calculate 3d edges:
4, = 45 550), B, = B@S. ‚53 ;0°.), i=1..N. (15)
Ji Yi Ji Yi Ji Yi
Note that weight of A,B, is W, =W .([@5 a0". ).
UF UF Ji y;
3 | GAUSS PYRAMID USAGE
Here we want to say a few words about related topic, namely Gauss pyramid usage for fast and robust calculation. It
should be noted that application of Gauss pyramid allows one to construct scale-independent algorithms [Poul S. Wu,
Ming L,, 1997].
A AN
*\ toplevel toplevel // {:\
EN /L
X
Let Lis ik, be Gauss pyramid of the “source” / Y
/ |
image: i? zi? is the bottom level of pyramid, i is the I=L
top level, (65 ) = i s: ) d and (5 ) = |: (5 jJ. }
/=1..L, And let li? (x, Wik be appropriate Gauss
pyramid of the “destination” image. The count of layers
L +1 should be chosen so that character size of the object
on the top level images i? and i7 is not less than several
pixels. Note that in practice Z —1 is usually equal to 2 +
4 bottom level bottom level
Figure 6. Gauss pyramids of "source" and
As soon as we construct Gauss pyramids we can apply the "destination" images
technique described in previous section to top-level
images i? and ij . As a result we obtain set of 3d edges Ja pi^ . Then, these 3d edges we use to construct
predictions for the lower layers:
(a )- = (a s ar (4 Bl ) 6; »" = 6 S L (4} ‚Br ) de
76 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
mM 65 —— t5 (F5 (C5 LI "m
"m
5 4