PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2017
87
Shadow Identification in High Resolution Satellite
Images in the Presence of Water Regions
Y. Mostafa and A. Abdelhafiz
Abstract
Shadow is a main obstacle in features extraction from high
resolution satellite images. Water areas provide low reflectance;
therefore, they are commonly classified as shadow. In tradition-
al shadow identification procedures, shadow is identified from
image histogram in which, water and shadow pixels are mixed.
In this study, a new image index is presented. The index
histogram separates non-shadow, shadow, and water pixels
using a proposed threshold method. The developed index is
then integrated into a complete approach for shadow identifi-
cation. Five study areas including water, shadow on water, and
shallow water regions are tested using the developed approach.
Afterward, an accuracy assessment is made to demonstrate
the approach efficiency. The experimental results show that
the proposed approach achieved overall accuracy about
ninety-three percent. The rate of detecting shadow on water is
about seventy-four percent. On the other hand, shallow water
is still misclassified as shadow.
Introduction
Shadow occurs when an object partially or totally occludes
direct light from a source of illumination. In remote sensing
images, pixels that present in shadow area appear darker than
usual. These dark pixels lead to obviously defected regions
that disturb the classical techniques of image analysis and
cause loss of surface information. Consequently, tasks like im-
age interpretation, objects detection, objects recognition, and
change detection have become more difficult or even impossi-
ble (Dare, 2005). On the other hand, shadow has its benefits as
it provides a considerable amount of useful information about
shape, relative position, and surface character of the object
generating shadow (Singh
et al
., 2012).
Early research in shadow detection have focused mainly
on terrain-cast shadows in mountainous areas (e.g., Rich-
ter,1998; Giles, 2001), and/or cloud shadows (e.g., Simpson
and Stitt,1998; Simpson
et al
.,2000; Wang
et al
.,1999) from
low and medium resolution imagery (e.g., Landsat MSS and
AVHRR data). Recently, with the wide availability of high
spatial resolution imagery (e.g., Ikonos and QuickBird data),
there is an increasing interest in ground feature shadow de-
tection (Zhou
et al
., 2009).
Tsai (2006) compared the effectiveness of various color
models for shadow detection, including hue-saturation-in-
tensity (HSI), hue-saturation-value (
HSV
), hue-chroma-value
(
HCV
), luma-in phase-quadrature (
YIQ
) and Luminance - chro-
ma blue - chroma red components (
yc
b
c
r
) models. Gevers and
Smeulders, 1999 proposed a color model (C
1
C
2
C
3
) represent-
ing the first, second, and third chrominace respectively. The
C
3
band from the color invariant indices C
1
C
2
C
3
has been used
for shadow detection by Arevalo
et al
. (2006) and Sarabandi
et
al
., (2004); then texture filters and region growing techniques
are applied to extract shadows. Cai
et al.
(2010) suggested
three different indices for shadow detection based on HSI col-
or space, which achieved more accurate results than using HSI
color space only. However, when (Wan
et al
., 2012) applied the
previous algorithms, shadows were mixed with water bodies.
Ma
et al
. (2008) presented an approach based on (
HSV
)
color space to detect shadow regions of buildings. The results
of detection revealed effective extraction of shadow area. Nev-
ertheless, the approach was not able to differentiate between
dark objects, such as water and shadow regions.
The previously mentioned classical shadow detection
methods were not able to distinguish between water and
shadow regions due to the similarity of the spectral values of
their pixels. Chen
et al
. (2007) proposed five indices to sepa-
rate shadow from water. The fifth index is recommended and
named as Spectral Shape Index (
SSI
). However, when Wan
(2012) examined the fifth index, many of non-shadow features
were extracted. After examination of the other four indices,
he found that the second index separated shadows from water
much better than the recommended index. Latest shadow
detection methods with their applications can be reviewed in
Namrita and Maxton (2014) and Shahtahmassebi
et al
., 2013.
In this paper, a new image index is developed. This index
serves in distinguishing between shadow and water regions
through a complete approach for shadow detection. Problem
definition and solutions are given in the next section. The
study area is shown in the third section followed by the pro-
posed approach. Next, an accuracy assessment of the results
is presented then validated through four cases of study. Final-
ly, conclusions are presented in the final section.
Problem Definition and Solution
Shadow can be identified by employing different types of
color models such as
IHS
,
HSV
,
YIQ
,
yc
b
c
r
and C
1
C
2
C
3
. In the
traditional approaches for shadow detection, the histogram of
a certain band from the employed color model is computed.
Shadow is identified as dark pixels, therefore the histogram
threshold method for shadow identification has been success-
fully used to achieve the decision surface in many previous
studies (e.g., Dare, 2005). From the histogram, image pixels are
classified into shadow parts and non-shadow parts according
to pixels darkness. In case of water area existence, water pixels
(which are dark pixels) appeared in the same region of shadow
pixels in the image histogram of all bands (red, green, and
blue) as shown in Figure 1c, 1d, and 1e). The histogram iden-
tifies shadow with accepted accuracy in c0ase of land only
without water regions existence (see, Tsai 2006). The obtained
accuracy depends mainly on the considered method to assign
the threshold value. Others developed further steps to obtain
Y. Mostafa is with the Civil Engineering Department, Faculty of
Engineering, Sohag University, Egypt (
).
A. Abdelhafiz is with the Civil Engineering Department, Faculty
of Engineering, Assuit University, Egypt (
.
Photogrammetric Engineering & Remote Sensing
Vol. 83, No. 2, February 2017, pp. 87–94.
0099-1112/17/87–94
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.2.87
