hul 2004 
ribe this 
ided, an 
rate the 
unique 
[Baarda, 
justment 
exclude 
al set of 
je data 
lometric 
unique 
ibed the 
'ometric 
3 will 
nts and 
cclusion 
cuss the 
; called 
will be 
scopical 
| terrain 
| to this 
/ height 
)perator 
ints or 
)perator 
tain the 
s likely 
"loating 
Plane". 
ind the 
it range 
ane" to 
rovided 
Similar 
)6]. 
[inimal 
] plane 
eans of 
set of 
'e must 
terrain 
terrain 
Z) are 
e. The 
rage is 
ndidate 
lue of 
image 
Among 
1didate 
image 
in the 
Jed. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
2.3 Unique Orthoimage Patch from Image Fusion 
When an optimal set of orthimage patches is found from all 
candidate sets. The subsequent task is to fuse the radiometric 
information from multiple orthoimage patches into unique one. 
As Fig. 1 shown, multiple orthoimage patches corresponding to 
a terrain area means multiple measurements corresponding to 
same ground locations. As the preceding section mentioned, a 
major orthoimage patch is decided in an optimal set of 
orthoimage patches. If we supposed the radiometric difference 
between this major orthoimage patch and other orthoimage 
shoubd be a constant. Meanwhile those differences should 
conform to the theory of normal distribution. But due to some 
factors in imaging process, e.g. sun location, atmospheric 
interference, terrain surface material, terrain occlusion and so 
on, some radiometric differences could not meet this 
assumption. That means that occasionally large random errors, 
i.e. blunders, will occur. When blunders exist, a least-squares 
adjustment may not be possible or will produce poor or invalid 
results. Therefore, in this study, Data Snooping method will be 
utilized to exclude the inappropriate radiometric differences 
during fusing the multiple orthoimage patches into a unique one. 
2.3.1 Data Snooping: Data Snooping was proposed by 
Baarda [1968] for blunder detection. In this study, this method 
is used to isolate the large radiometric difference between two 
orthoimage patches corresponding to the same groundel at some 
confidence level. The detailed derivation of this method can 
also be found in [Wolf and Ghilani, 1997]. In this study, the 
adjustment of radiometric difference between two orthoimage 
patch can be expressed in matrix form as 
L+V =AX (1) 
where AT = [1 ] 
X= fx, is the estimated radiometric difference parameter 
vector. IZ = [/ ] 
of radiometric difference between two orthoimage patches. 
KE =ir vw 
observation is regarded as the same weight, therefore, Eq.(1) 
d is the coefficient matrix, 
Li, is the observation matrix 
vla; is the residual vector. The 
; ol 
has a covariance matrix W — Sy 0, . 
n*n 
According to [Wolf and Ghilani, 1997], the relation between the 
residual vector and the true error vector can be expressed as the 
following form 
E=-0,W6 (2) 
where 
o. = w^ — AO, À" = w^ 0, 
0. (AWA 
Now consider the case when all measurements have zero errors 
except for a particular observation / i which contains a blunder 
of size Al ; - A vector of the true errors, AE , can be expressed 
Aee[0 0 — 9 A 0 . 9l 
S 
gEALO: «07,0 495.9] 
If the original measurement are uncorrelated, the specific 
a 
correction for Av, can be expressed as 
Av; = -quW; A; = -r, M, G) 
where {;; is the ith diagonal element of the 0. matrix, W; 
is the ith diagonal term of the weight matrix, W , and 
F— QW; 
; 1$ the observational redundancy number. 
From Eq.(3), the V; to an observation can be calculated and 
used to isolate measurement blunders by computing the 
standardized residuals from the diagonal elements of the 0. 
matrix as 
Vi V; 
Wi = TE 
= LU 
= 
Tyan Ton Gi iO ATi 
where W, is the standardized residual, V; the computed 
(4) 
,, matrix, O' is the 
residual, q'; the diagonal element of the Q 
known unit weight standard deviation. When the O, is 
unknown, the /; test statistic can be defined from Eq.(4) by 
replacing O', with 
  
Vi 
f; eM b V (5) 
TO Ti 
As Eq.(5) defined, where 
5 
G, = eee (PH = W,V; ). 
n—u-1 Y 
/ 
The approach is to use a rejection level given by a / distribution 
test with n-u-/ degree of freedom. The observation with the 
largest absolute value of /; as given by Eq. (5) is rejected when 
it is greater than the rejection level. 
2.3.2 Image Fusion: After the greater radiometric difference 
between each orthoimage patch and the major orthoimage patch 
corresponding to all groundels are excluded, the final 
orthoimage patch should be fused from this optimal set. The 
concept of data fusion could be utilized in this step. The 
original definition of data fusion is to fuse the data from 
different sensors. Although the orthoimage patch data are from 
the same kind of sensor, aerial camera, the idea could be used 
without violation. There are many approaches to dealing with 
this problem. [Jin ef al., 2002] In this study, the simplest 
approach is employed to fuse the radiometric information into 
unique one from multiple information. Namely, the major 
orthoimage pacth is used as basic radiometric information. Then 
radiometric information that is not rejected in each groundel are 
summed up and averaged to obtain the final radiometric 
information in each groundel. If only one radiometric 
information is obtained from basic radiometric information, 
then the radiometric information of this groundel will be null. 
3. INVESTIGATION INTO THE QUALITY OF FINAL 
ORTHOIMAGE PATCH 
This section discusses the quality about the generated 
orthoimage patch. The whole processes of orthoimage patch 
generation from aerial photos can be simple separated into the 
following steps: (l)the imaging, (2).the digitizing, (3).the 
determination of the exterior parameters of aerial images, 
(4).the acquisition the surface elevation, (5).the rectification 
process, and (6).the output. 
Although many steps will affect the quality of final 
orthorectified image patch, the camera quality definitely have 
877