340 
ı urban 
extent 
as pose 
10lding 
y other 
in the 
mented 
lineate 
atically 
regions 
olding 
"tween 
e and 
terrain 
urface 
12001, 
nodels 
ogram 
of the 
rs, sun 
red to 
1as not 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
3. PRESENT APPROACH 
The present work utilized the mean-shift analysis 
(Fukunaka and Hosteler, 1975; Cheng, 1995; Comanicu 
and Meer, 2002) filtering and segmentation for shadow 
region extraction from high-resolution air-photos. The 
lowest mean lightness parameter added to the segmented 
image finally will automatically extract the shadow 
regions. Since shadow regions are the darkest regions, 
applying the lowest mean lightness value to threshold the 
segmented image will extract shadow regions. 
Panchromatic (PAN-grey level) was used as input. ADS 
40 NIR image covering a part of northern part of Tokyo 
(figure 1) was tested and significant results obtained. 
DN AN U 
   
The lowest lightness value can be attributed to the darkest 
shadow regions (especially in NIR images), and shadow 
regions will be extracted by using Boolean operation. By 
image arithmetic operator, the extracted shadow regions 
were eliminated from the original image and texture of the 
shadow regions was separated. The texture information 
under shadow regions was enhanced by using the adaptive 
contrast enhancement (ACE) techniques suggested by 
Stark (2000). 
4. EXPERIMENTAL RESULTS 
The filtered and segmented ADS40 image is shown in figure 2. 
All the obtained segmented images were thresholded by using 
the mean lightness thresholding that separated shadow regions 
and non-shadow regions (figure 3). From the figure 3, it was 
809 
observed that all the dark pixels are not shadow regions as some 
of them represent water bodies (A in figure 3), dark coloured 
roofs (B in figure 3), newly paved tar roads and similar objects. 
Figure 4 shows the texture information in the shadow regions. 
The adaptive contrast enhancement helped to enhance the 
shadow regions and finally an integration of the shadow texture 
regions with the shadow-removed image resulted in a 
shadow-free image (figure 4). 
  
Figure 3. Extracted Shadow regions and false positive 
regions (circled areas A- water bodies; B-dark coloured 
roofs) 
  
Figure 4. Texture information under shadow regions