Performance Evaluation of Object-based and
Pixel-based Building Detection Algorithms
from Very High Spatial Resolution Imagery
Iman Khosravi, Mehdi Momeni, and Maryam Rahnemoonfar
Abstract
This paper reviews and evaluates four building extraction
algorithms including two pixel-based and two object-based
methods using a diverse set of very high spatial resolu-
tion imagery. The applied images are chosen from different
places (the cities of Isfahan, Tehran, and Ankara) and dif-
ferent sensors (QuickBird and GeoEye-1), which are diverse
in terms of building shape, size, color, height, alignment,
brightness, and density. The results indicate that the per-
formance and the reliability of two object-based algorithms
are better than pixel-based algorithms; about 10 percent
to 15 percent better for the building detection rate and 6
percent to 10 percent better for the reliability rate. However,
in some cases, the detection rate of pixel-based algorithms
has been greater than 80 percent, which is a satisfactory
result. On the other hand, segmentation errors can cause
limitations and errors in the object-based algorithms, so
that the commission error of object-based algorithms has
been higher than pixel-based algorithms in some cases.
Introduction
Building detection from very high spatial resolution (
VHSR
)
imagery has been an active research topic over the past few
years. Up to now, several building detection algorithms have
been proposed in the literature which can be divided into two
categories: pixel-based methods and object-based methods. In
the first group, only pixels and often their spectral attributes
are used (Hester
et al.,
2008). On the other hand, the process-
ing unit of the second group is an object, where the non-spec-
tral attributes of the objects (such as proximity and geometry
attributes) can be used in addition to their spectral attributes
(Chubey
et al
., 2006).
A strategy of some building detection methods in the first
category uses only the clustering methods (Wei
et al
., 2004).
The combination of clustering and segmentation methods is
used in some papers (e.g., Hai-yue
et al
., 2006; Jiang
et al
.,
2008; Ghanea
et al
., 2011). There are some pixel-based meth-
ods which use the combination of spectral and morphological
indices such as differential morphology profile (
DMP
) or the
morphological attribute profiles (
AP
s) (Jin and Davis, 2005;
Mura
et al
., 2010; Huang and Zhang, 2011; Huang and Zhang,
2012). Some papers have used only morphological methods
(Meng
et al
., 2009). Another strategy of some pixel-based
methods is the combination of spectral indices, clustering and
morphological methods (Aytekin
et al
., 2012).
In the second category (object-based methods), the most
important step is segmentation (Blaschke, 2010). Benz
et al
.
(2004) and Taubenbock
et al
. (2010), presented an object-
based, multi-level hierarchical classification method based on
a multi-resolution segmentation to detect the urban features.
Bouziani
et al
. (2010) detected building by a rule-based
classification method based on an automatic region growing
segmentation. More recently, an edge-based segmentation
has been used in some papers (e.g., Kanjir
et al
., 2008; Hu
and Weng, 2011) to classify urban features especially build-
ing regions. Meng
et al
. (2012) presented a hybrid approach
of object-based and morphology-based methods to detect
residential buildings from lidar data and aerial images. In a
recent paper, a classifier ensemble strategy based on combin-
ing pixel-based and object-based processing is presented to
detect urban features (Huang and Zhang, 2013).
Up to now, a few papers have compared the pixel-based
and object-based analysis for classifying urban features espe-
cially buildings (Wang
et al
., 2007; Cleve
et al
., 2008; Myint
et al
., 2011). Therefore, the present study aims to compare
pixel-based and object-based analysis for detecting build-
ing. For this purpose, four building detection algorithms are
reviewed which the first two algorithms are pixel-based and
similar to the works proposed by Ghanea
et al
. (2011) and
Aytekin
et al
. (2012) and the rest of the algorithms are object-
based approaches performed by
eCognition
®
Developer
and
ENVI Feature Extraction
software. This paper has tried to use
a diverse set of
VHSR
images for comparing these algorithms.
The applied images are chosen from different places and
two different sensors, i.e., QuickBird and GeoEye-1; they are
diverse in terms of building shape, size, color, height, align-
ment, brightness, and density.
The reminder of paper is organized as follows: In the next
Section, the data applied to this paper is explained, followed
by the four aforementioned algorithms. Then, the result of each
algorithm applied on the dataset and comparison between
their performances is presented, followed by our conclusions.
Iman Khosravi and Mehdi Momeni are with the Department
of Surveying Engineering, Faculty of Engineering, University
of Isfahan, Isfahan, I.R. Iran (
.
Maryam Rahnemoonfar is with the School of Engineering &
Computing Sciences, Texas A&M University-Corpus Christi, TX.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 5, June 2014, pp. 519–528.
0099-1112/14/8006–519
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.6.519
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2014
519
