International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 Intei 
image, because the conjugate point simply does not exist. | The 
Another problem is the ambiguity of object structures, because | area 
of the various possibilities for correlation in some cases false | x-di 
points are matched. Furthermore, in areas of low texture, mat- case 
ching homologue points becomes also impossible, because dif- | 
ferent positions in these areas can hardly be distinguished. | But, 
Trar 
The mentioned problems are independent from the image | segr 
acquisition system. Thus, the new approach has to deal with | poin 
similar problems as they are common in digital photogramme- | coul 
try. These problems are very much dependent from the objects who 
properties, illuminations conditions, and from the structural | spor 
image properties as well. To a certain extent improvements can | wer 
be achieved by modifications of the data acquisition their 
parameters, as e.g. the base-height ratio in conventional | thod 
photogrammetry. | sign 
rem: 
The second possibility for generating a true orthoimage is the | ture 
segmentation of corresponding regions by means of automatic | men 
or manual procedures. It is evident, that also in this case the | but 
success is very much dependent from image characteristics. The | ges 
application of automatic segmentation methods can only be ef- | metl 
fective, if the segmentation algorithms yield quasi identical seg- | 
ment contours in both image data sets. This can be achievable if | The 
roofs from buildings or other distinct geometric elements of proc 
man-made objects are concerned. In more complex situations, | mair 
the image segmentation approach might fail. Detailed experi- | high 
mental studies and methodological improvements are mati 
necessary. | mod 
gene 
For a preliminary study, the manual segmentation was selected | seve 
as described in the following. For this purpose, HRSC-AX | lutio 
pushbroom scanner images were used, which show a small part obje: 
of a residential area, acquired in 4500 m flying altitude (fig. 5). | men 
This part was imaged in two directions perpendicular to each | - its p 
other. Hence, the requirements were fulfilled to generate a true digit 
orthoimage by means of the new approach. After the measure- | ratio 
ment of the area defining points in both datasets, in this case | quali 
roof of the buildings, the correct ground corner positions were 
calculated by combining the coordinates from the parallel pro- | How 
jected directions, in order to obtain the area locations in the true cont: 
orthoimage. If coordinates of the area defining points exist in | corre 
the two image strips, segments of the original image data can be | imag 
mapped to the aimed true orthoimage, utilizing an appropriate sour 
transformation. For this purpose, the bilinear transformation | the 
was used, which is particularly suitable and practical for recti- | Rem 
fying flat quadrangular areas, such as roofs etc. Such areas conte 
occur by a segmentation of any area that could be divided into | scani 
triangles and quadrangles. gaps 
imag 
Figure 5 shows transformed roof areas of the buildings in their | Filles 
correct position, based on different datasets imaged from diffe- | priate 
rent directions. An entire roof was corrected, if it was com- | The 
pletely visible in the image. In order to illustrate the manual | tradit 
segmentation in principle, the roofs of the imaged buildings 
were measured separately and the remaining features, e.g. tree- | Furth 
tops, were ignored. Due to the fact, that the roof areas were | usabi 
transformed into their proper ground position, gaps with blank sitior 
content remained. This necessitates to fill them from the appro- | turing 
priate image. A correction of the radiometric variance and | acqui 
brightness values in the filled areas has to be calculated, due to | imag 
the current demands on a true orthoimage. Anyway, it was not Figure 5. At the top, part of a residential area imaged by HRSC- 
determined in this example, because of demonstrating the sche- AX in x-direction. In the middle, corrected image in 
matic correction by means of the manual segmentation method. y-direction, based on a dataset imaged in x- | 
Hence, the correction of roof areas in y-direction in an x-direc- direction. Below, corrected image in x-direction, 
ted overflown image is well visible, in the middle of figure 5. based on a x-directed overflown dataset. 
696