= p "n 2:
V = 2 (A,-Bi) =llc’ll =(||Al| sin §) =min (21)
1,5
ea Er [0| =min (22)
It is eauivalence to correlation coefficient (Fig. 2).
The difference betwen A and B can also be defined by the
absolute norm of vector C, i.e.
pim 321A; Bi 321€; Lo min (23)
Fig.3 shows that all points along square sides 1234, which
has symmetric centre at point A and diagonals paralle to
axis u or v, is equal in norm P.
Analysis of the algorithms with derivative preprocessing is
more complete. However for finding the discrepancy among the
criterions only, we may take a simple example to make easy.
Considering the sum of absolute differences with first de-
rivative processing, the following expression can be derived
from (10.1)
"n-1 n-]
D,72 ZA: Ar) (Bi Bi0] 2 2214; B2 Art Bin)
i21 i21
-$316,-6,.l -E3l4C;l o min (24)
It will be seen from Fig. 3, that points along the line lo
passing through point A and spreading 45'angle from axis u,
have \aCu|=0; and all points on line ltparallel to le, have
làcuv'lconstant, that is in proportion to the distance from
l4 to lo.
Fig. 4 Discrepancy of the criterions
The expressions (15),(19), (180, (222, (24) demonstrat various
criterions in the same vector space and shows all the crite-
rions are not identical.When the target vector A is given and
several match vectors (such as I,1.IL,IV.V shown in Fig.4) ap-
pear in a sertain searching area of the right image, the dif-
ferent criterion may result different correlation. For ins-
tance, vector 1 may be selected as the correspondence of A by
correlation coeffient criterion and vectors I.IX.1V or V may
- 419 -