International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
important influence on the quality of aerial images and 
aerotriangulation is largely concerd with the quality of exterior 
parameters. However, in this study, 
aerotraingulation and the quality of aerial images are supposed 
to be sufficient for providing good conditions for 
orthorectification. Then, the quality about the final orthoimage 
patch can be discussed more simply. Subsequently, let's discuss 
the quality of final orthoimage from the geometric quality and 
radiometric quality. 
3.1 Geometric Quality 
If the lens distortion and the atmospheric affection can be 
neglected, the geometric quality are concerned with 
(1).scanning accuracy of aerial photos, (2).the transformation 
accuracy between photo coordinates and image coordinate, 
(3).the accuracy of exterior orientation parameter from 
aerotriangulation, (4).the 3-D coordinate accuracy of this terrain 
area, and (5).ground sampling distance (GSD). Of course, the 
factor (3) is completely affected by the accuracy of ground 
controls and the aerotriangulation. If the aerotriangulation 
accuracy completely meets the requirements, the major factor 
should be from the determination of the 3-D coordinates no 
matter what it is identified by the operator interactively or by 
image matching automatically. Shadows and occlusions are two 
great factors to seriously affect the identification and 
determination of 3-D points. 
3.2 Radiometric Quality 
If the geometric quality is sufficient, the radiometric quality 
depends on (l).terrain material, (2).sun location, (3). 
atmospheric condition, (4).camera quality, (5).resampling 
algorithm as well as (6).image balancing and enhancement 
techniques. The shadow from adjacent terrain areas is one of the 
most serious factor that affect the radiometric information. 
Another serious factor is occlusion effect from other adjacent 
higher building. These are two different topics to reconstruct 
the radiometric information from shadow and occlusion area. 
From the above discussion, shadows and occlusions are two 
serious problems when producing orthoimage patches. 
Therefore, the choice of orthoimage patches should take these 
two important factors into consideration, especially for control 
information in aerotriangulation. By selecting the local highest 
terrain, these two problems can be avoided. 
4. EXPERIMENTS AND DISCUSSIONS 
The photos used in this study were taken by 304.921000 mm 
focal-length aerial camera in 2000. The photo scale is about 
1/5,000. The scanning pixel size is about 15pm, which is about 
7.5 cm ground resolution. 
According to analytical aerotriangulation, X and Y coordinates 
of pass points can quite routinely be located analytically to an 
accuracy of within about 1/15,000 of the flying height, and Z 
coordinates can be located to an accuracy of about 1/10000 of 
the flying height.[Wolf and Dewitt, 2000] Therefore, the 
spacing in X, Y, and Z coordinate components are set to 0.15m, 
0.15m, and 0.20m according to its reachable accuracies. And 
the range of height approximation is supposed to it's true height 
+/- 2m. The true height is determined by multiple image space 
intersection in this test. 
From Table 1, it is shown that a total 17 orthoimage patches 
were generated by this proposed methodology. The number in 
parentheses after each point number indicates the number of 
related aerial images used to generate the orthoimage patch. 
the accuracies of 
878 
Except one concrete floor and one bridge surface, the selected 
horizontal terrain areas are almost classified into 3 kinds: 
buildings, roads, and basketball courts, as shown in Fig. 2. 
Now, let’s examine the generation of building orthoimage patch 
75550677 from Fig. 3. The optimal set of orthoimage patches, 
composed of 3 related orthoimage patches, are generated by 
minimal radiometric difference, described in Section 2.2. In this 
case, orthoimage patch shown in Fig. 3B is sclected as the 
major orthoimage patch. Then Data Snooping was used to 
exclude possible gross errors during image fusion. Obviously, if 
Data Snooping was not employed, the result of image fusion by 
simply averaging will be affected by the occlusion area, as seen 
the areas in black rectangular in Figs. 2 and 3. Another obvious 
example is the area in the white circles. A streetlight is imaged 
in this area. Its imaging locations were different because of the 
different aerial camera location. Data snooping remove the 
image information out (sce Fig. 2) successfully. The excluded 
image information is given by grey level 0 or 255 in this study. 
As Fig. 2C shown, the areas in white circles are the image of 
basketball stands. Their image information was also excluded 
by Data Snooping. These three cases can clearly state the 
effectiveness of Data Snooping used before the process of 
image fusion. 
  
  
C. Orthoimage patch of Basketball Court terrain (1261375) 
Figure 2: Different terrain area selected as orthoimage patches 
As concerned with geometric quality, sharp locations were 
selected in each orthimage patch for the evaluations of 
geometrical accuracies. The 3-D coordinates of each sharp