AUTOMATIC EXTRACTION OF SHADOW REGIONS IN HIGH-RESOLUTION ADS40 
IMAGES - By Robust Approach of Feature Spaces Analysis 
B. Babu MADHAVAN*, Kikuo TACHIBANA, Tadashi SASAGAWA, Hiroyuki OKADA and 
Yasuke SHIMOZUMA 
Geographic Information Systems Institute, PASCO Corporation, , Tokyo, Japan 
Email: mb pasco.co.jp 
KEY WORDS: Vision, Urban, detection, extraction, recognition, feature detection, automation. 
ABSTRACT: 
Due to high object density and high proportion of shadow-covered areas, it is usually quite difficult to extract information in urban 
high-resolution airborne images from Leica Geosystems' airborne multispectral line sensor, ADS40. The shadows of various extent 
cause problem in image matching for elevation extraction, and obstruct road extraction. Further, the occluded and shadow areas pose 
problem to the interpretability of orthophotos. This paper describes strategy of robust feature space analysis and multilevel thresholding 
adopted to extract shadow regions in ADS40 images automatically without the aid of Digital Surface models (DSM) or any other 
accessory data. 
In this work, low-level computer vision tasks for shadow extraction, elimination, and enhancement of texture information in the 
shadow regions in high-resolution ADS 40 digital aerial images were presented. A general nonparametric technique was implemented 
for the ADS 40 data points in the joint spatial- range domain for the analysis of a complex multimodal feature space and to delineate 
arbitrarily shaped clusters in it. The filtered image was segmented and then subjected to a bimodal thresholding to extract automatically 
shadow regions. Using an adaptive contrast enhancement approach, texture in the shadow regions was enhanced. All shadow regions 
caused by buildings, trees and other smaller objects in the urban areas were robustly extracted. 
2. PREVIOUS METHODS 
1. INTRODUCTION 
Shadows were detected through simple thresholding 
While shadow detection and analysis in high-resolution (Nagao et al. 1979), by using correspondence between 
digital air-photos is difficult, the development of shadow edges and the geometric edges (Lowe and 
techniques to extract, eliminate and enhance are still Binford, 1981), shadow simulation by using digital terrain 
continuing. It is usually quite difficult to extract models (DTM) (Stevens et al 1995); using digital surface 
information in urban high-resolution airborne imageries models (DSM) and sun angle parameters (Xu and Li 2001, 
due to high object density and especially high proportion Nakajima et al 2002), DSM and digital building models 
of the shadow-covered areas. Images from multi-lines (DBM) (Rau et al., 2000); and by colour histogram 
scanner technology based Airborne Digital Sensors such matching (Gerke et al, 2001). Thus in most of the 
as ADS40 from Leica Geosystems have more shadow previous attempts, DSM, DBM, sensor parameters, sun 
areas especially in urban region. So, it becomes difficult elevation and zenith angle parameters are required to 
to extract the information in the shadow-covered area. extract shadow regions. But the present approach has not 
considered any of these key factors. 
808 
* Corresponding author 
Intern 
(Ful 
and | 
regio 
lowe 
imag 
regio 
appl; 
segir 
Panc 
40 N 
(figu 
Fi 
imag 
wert 
shad 
und 
cont 
Star 
The 
All 1 
the ı 
and