3. Istanbul 2004
n or information
Rau et al.; 2000).
WwW imagery at the
cquisition. That
from the Ikonos.
ransformation to
) compensate the
ompensation are
he sun and sky
on a hierarchical
mentation of the
(mean value and
nd orientation),
s in sun side and
overy the method
pouring segments
nformation under
ing section.
lysis, so the first
on to produce
alculation of all
deviation), form
orientation) and
Sun azimuth at
acquisition time
À
TS
>
detect shadow is
and threshoid to
segments darker
yments with grey
ined as potential
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
2. Local histogram analysis and the form attributes are
used to refine the shadow detection for discriminating
the projected shade, the self-shadow and penumbra.
The form criteria used are the area, the length/width
and the compactness.
The neighbour shadow segments are merged to form a
more geometrically significant segment. Then, the
orientation of each side is calculated and compared to
the sun azimuth direction. If the difference is lower
than a fixed tolerance level, the segment is regarded
as having one or more sides directed according to the
azimuth of the sun and increases its chance to be
shade.
4. Each segment detected as shadow, must be confirmed
by analysing the form of its neighbouring segments in
the sun side. If there is two or more segments, they
merged if their mean and variance are closed to get
meaningful segment. For a regular form segment, it
can be a building and confirm the shadow detected. In
other case, the analysis of the under-segments can
confirm the shadow by the presence of self shadow,
projected shade and penumbra.
o3
3.2 Methodology for information restitution from shadow
The information retrieval under the shade is based on texture
invariance by shadow. Considering that a surface texture does
not significantly change when shadowed. So to retrieve a
surface in shadow, we can use a contextual texture analysis
between the shadowed segment and its neighbouring segments
in the shadow side. Because surfaces receiving shadow from an
object are located at the opposite side of the sun, they are
located in the shadow side.
The principal stages of the methodology presented in figure 2
are:
1. For each segment shade, the list of its neighbour in
shadow side is checked. If a neighbour segment is a
shadowed segment, it is excluded. Then, the textural
attributes of the segments in that list are bring out.
2. Analysis of texture between the shadow and all the
segments of the shadow side vicinity list. For this
analysis, we calculate the difference of texture
between each neighbouring segment of the list and
the segment shade. The segment having a texture
difference close to zero is considered as having the
same texture with the shadowed segment. And thus
two segments are from the same surface.
3. The surface type represented by the neighbour
segment with the same texture as the shadow is
identified from a soil occupation map or a
classification result. So the segment in shadow is
allowed to the same surface.
For the shadow effects compensation, the gamma
transformation as formulated in the equation eq (1) was used:
1/5
OutPixel = 2047*(InPixel/2047) "(5
Where: OutPixel : Pixel value after correction
Inpixel : shadow pixel before correction
S. parameter of transformation
The parameter of transformation & is calculated for each
shadow area using the shadow segment mean value and its
neighbour in sunlight representing the same surface.
Using this parameter, each pixel value in the shadow is
corrected using the formula (1) and its new value represents the
value that the pixel must have if he isn't in shadow.
Original Image segments and Shadow segments and Sail occupation
Image textural attribute contextual and textural map
attributes
v Textural features analysis - T
between shadow and its vicinity
i on shadow side -
we ad uU >
Pe Surface in shadow m
deduction (surface with tbe d Pe Validation >
Ceu same texture like shadow lt meme Dieu
re tt nnt De.
^ /^ Correcting parameters X
n : M /
(Shadow pizel value a calculation for each ]
en shadow segmert Let
<< compensation
Se, rn
Figure 2: Diagram for information recovery and de-shadowing.
4. DATA AND SITE OF STUDY
4.1 Site study
The site study area is the town of Sherbrooke, selected for the
availability of IKONOS images. Two sites were retained for
their different characteristics. The first site is the Western
Campus of the University of Sherbrooke, with large buildings
for collective use. The density of the buildings is low and the
presence of shade is well highlighted beside the various
buildings. The second site is selected in the town centre, with
buildings of various sizes and of different use (trade, utility
services, residences, etc). The density of the frame on this site is
rather strong.
4.2 Data characteristics
Ikonos images (panchromatic and mutlispectral) covering the
city of Sherbrooke were acquired on May 20, 2001. We retain
the panchromatic image for this study, because it is the most
affected by shadow effects. So, the panchromatic image
covering these two sites were extracted for testing the methods.
The image was acquired by a very clearly atmosphere and do
not need atmospheric correction. The image are georeferenced
for calculating the shadow sides orientation to be compared to
the sun azimuth. There is no expressed need for geometric
correction for testing the shadow detection method and the
restitution of information under shadow. An small extract of the
panchromatic Ikonos image from the Sherbrooke University
campus is presented at figure 3.
Some ancillary data are provided: The positions of the sun and
the sensor (azimuth and rise) at the acquisition time to compare
with the shadow sides orientation, the land use map to validate
the information under shadow restitution. An shadow map
derived by image interpretation is also provided to validate the
shadow detection results.
