Istanbul 2004
with rational
002 Annual
, April 19-26
Jata Products
Centre.
' Handbook
2.
an integrated
Sensors and
14th EARSel
panese Space
the “Instituto
RESTITUTION OF INFORMATION UNDER SHADOW IN REMOTE SENSING HIGH SPACE
RESOLUTION IMAGES: APPLICATION TO IKONOS DATA OF SHERBROOKE CITY
; Amani Massalabi, Dong-Chen He, Goze B. Bénié et Eric Beaudry
Equipe ESTRITEL, Centre d'applications et de recherches en télédétection (CARTEL).
Université de Sherbrooke
Sherbrooke, QC, JIK 2R1
E-mail : Massalabi.Amani(@Usherbrooke.ca
KEY WORDS: Restitution, Spatial High Resolution, Urban, shadow, urban, Ikonos.
ABSTRACT:
Shadow regions are very present in the high spatial resolution images, particularly in urban environment. Shades have negative effects.
They strongly disturb the classical techniques of image analysis by modifying surface appearance and sometimes involve loss of
information under the surface they covered. In this study, we try to restore the information masked under the shadow by finding the types
of surface covered by the shade. This search is based on the analysis of contextual, spectral and textural information from the shadow and
its vicinity.This restitution supposes that shades are well identified, which was the aim of our former work. It takes into account the space
configuration of the vicinity between the objects located sun side and surfaces in shade side which are supposed to receive the shadow.
The application on an IKONOS image of Sherbrooke show that it is indeed possible to restore the information laying under the shadow
and even discriminate several types of surface under the same shade, such as shades of buildings which are projected at the same time on
grass, a portion of parking or on another building. The method also makes it possible to correct the shadow effects on the images.
1. INTRODUCTION
The advent of the sensors with very high space resolution
(Ikonos, QuickBird, etc) opens a new era for remote sensing,
mainly for applications where the need for details is essential as
in urban environment. With thus images, we can distinguish
clearly detailed features such as building structure, roads,
vehicules, and trees (Nakajima et a/., 2002).
However, the enthusiasm of the users remains measured,
because of new difficulties in the analysis of these images.
These difficulties are the need of new techniques to extract
information from these data and the presence of undesired
features like shadow of clouds and high objects like buildings
and trees. Despite a suitable choice of the hour of acquisition,
shadow is very present on the images of very high spatial
resolution mainly in urban environment. The amount of shadow
increase with the spatial resolution.
The negative effects of shadowing enormously degrade the
visual quality of the images and cause several nuisances during
the analysis of these images by modifing spectral response of
the part of surface in shadow. Thus, surfaces under shadow are
often confused with other types of objects which have a same
spectral signatures as the shadow. The results are a
misclassification of area covered by shadow. All techniques of
information extraction from image are influenced by the
presence of shadow by modifing or masking information under
shadow. For a detailled mapping of urban area, using a very
high spatial resolution images, it is important to recover
information under shadow to complete the mapping. But for
recovering information in shadow, that shadow must be
accurately detected.
The object of this study is precisely to well detect shadow in
images and recover information from surface in shadow and
also to compensate their negative effects by de-shadowing
images.
Some techniques for shadow detection are developped in
moving object detection from videographic images (Prati et a/.,
2001); but, in remote sensing, only few works on shadow
detection are carried out mainly in builduing detection (Bruce
Irvin et a/., 1989, Chungan et a/., 1998) and clouds shadow
correction. Recently, some works on shadow detection on the
very high spatial resolution like Ikonos are published. Methods
use principaly color and spectral proprieties to detect shadow
(Adler-Golden et al., 2002). Some geometrical proprieties (area,
length, width, parallel sides, right angle, etc) are added to
improve the detection. Our shadow detection method is based
also on spectral and form proprieties, but include the sun
azimuth orientation at the acquistion time. Some contextual
proprieties (vicinity, relative position sun object and shadow)
are also added to improve detction or to confirm the results.
For the restitution of information under shadow, we use the
contextual and textural features between the shadow detected
and the neighboor surfaces in the shadow side. These surfaces
on shadow are supposed to receive projected shadow, even
surfaces on sun side are supposed to be objects generating
shadow. The main hypothesis is that the texture features are
shadow invariant and the neighbooring surface with the same
texture like shadow is considered to be the same surface under
that shadow.
After the surface under shadow is detected, negative effects can
be removed by transforming shadow pixels value. The
transforming parameters are calculated for each shadow zone
using it mean value and the neighboor segment of the same
surface.