MULTI-CRITERIONS FOR SIMILARITY ASSESSMENT
IN PHOTOGRAMETRIC IMAGE CORRELATION
Lin Zongjian
Wuhan Technical University of Surveying and Mapping
Wuhan, China
ABSTRACT
Several conventional similarity criterions in photogrammetric
image correlation are analysed by the use of vector algebra.
Theoretical imperfections of these criterions are discussed.
As a result, an algorithm with dual or triplex criterions for
stereo image matching is proposed. It has been verified by
theory as well as experiments, that the new method can give
better results than the methods with correlation coefficient
algorithm or with other single criterion, when both reliabi-
lity and speed of computation are considered.
1. INTRODUCTION
Similarity assessment plays an important part in photogram-
metric image correlation. Helava(1976), Dowman and Haggag
(1977) have made many investgations to eveluate the simila-
rity critérions. The results from their experiments shows
that every criterion has its advantages and disadvantages
different from the others. In general, the correlation coef-
ficient criterion gives best quality and takes longest com-
putation time.
In order to make a theoretical elucidation of the relations
and differences among the criterions and explore better paths
for similarity assessment, a method using vector algebra for
analysis of similarity criterions is accepted.
2. ANALYSIS OF THE CRITERIONS
In thise paper, the scope of image correlation/ or matching
is limited in photogrammetric field to seek the parallaxes
from the digitized stereo images. Several criterions applied
to conventional correlation algorithms can be briefly de-
scribed as following.
(a) Covariance
0,,2 & (0, — a) (b; — b) — max ( 15
i=1
Where a4 and b4 indicate the grey level with pixel number i
respectively in left image patch and in right image patch of
- 45 -