1 of small 
essing to 
ustness to 
automatic 
iciency of 
oach with 
> primary 
> potential 
t reached 
ical image 
rocessing 
the single 
istered to 
> identical 
nined, the 
sformation 
pairs has 
vever, the 
1. Various 
]t in SAR 
n function 
‚on, which 
ied on the 
10ise ratio 
d the 2D 
difference 
h a single 
lgorithms, 
procedure 
1e primary 
he 
rem 
r Surveying 
Mingsheng Liao 
normalized correlation coefficient algorithm, which tolerates any linear radiometric relationship between the images. 
Then the global matching procedure based on the bridge mode and probability relaxation technique is used to ensure the 
reliability of the matching result. The automatic strategy has been used for the registration between SPOT and TM 
images, which are acquired from different sensors (Zhang 1998). Finally, the Least Square (LS) matching algorithm is 
used to improve the precision (less than 1/10 pix.). The frequency independent coherence estimator (Guarnieri 1997) is 
used as the matching criteria of LS algorithm. The experiment is carried out successfully with SIR-C / L&C band data. 
2. METHODOLOGY 
21 Primary Image Matching 
The image matching is based on the measurement of similarity of image structures or gray level distribution between 
two images. At this scenario, only the intensity image of SLC data is used in image matching procedure. Firstly the 
feature points are detected with Forstner interest operator in the master image (Wang 1990), then the initial matching 
based on the similarity is conducted to find the candidates of the corresponding point in the slave image. The primary 
matching procedure employed normalized correlation coefficients between 0 and 1, which is defined as 
C( p.q) 
BTE = (D 
| sg 0 s Un 
Here C,, (p, q) and C,...(p, q) are respectively the variance of master image g(x, y) and slave image gY x, y) , C(p,q) is 
the covariance of g(x, y) and g'(x, y). The closer to 1 the value of p. the more similar images are to each other. The 
advantage of using the normalized correlation coefficient is that it tolerates any linear radiometric relationship between 
the two images. The computational complexity is not obvious. 
But the single point matching does not take the compatibility of matching results with its neighbors into consideration. 
It is difficult to improve the reliability for only the finite information contained in a matching window. In order to 
overcome the drawback, an algorithm of global matching with relaxation has been developed and further used to the 
second step of our registration procedure, which ensure the reliability of the matching result (Zhang 1990). 
22 Global Image Matching with Probability Relaxation 
As well known, image matching is used to define the corresponding point pair, that is the point j in shve image 
corresponding to the point i in the master image. Assume that there are an object set Oz( O,, O,, ..., O,) anda class set 
Cz(C, C,, ..., C, ], the pointi as an Óbject", the points jin slave image are classif ied into Class". If image matching is 
going to solve the problem of O;e C +» the global consistency should be considered. Therefore the probability of O;e C; 
and the compatible coefficient C(i,j; h,k) of O,e C ; ^ O,€ C, must be defined according to the probability relaxation 
algorithm. 
Since the area based matching algorithms use the similarity of gray level distribution as their measurement of image 
matching, the correlation coefficient p, may be used as the measurement of P; of O;e C;. Based on the principle of 
Correlation bebveen segments 
iH and RR. Pr Hs jk 
  
  
  
  
  
k e UTD 
Win | #15) 
resampling 
j 
n 1k — m 
E CA) * P8) 
i h 
Figure 1. Probability relaxation with bridge model 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 187