ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision‘, Graz, 2002 
  
INVESTIGATION OF THE MPEG-7 HOMOGENEUOS TEXTURE DESCRIPTOR FOR 
THE AUTOMATIC EXTRACTION OF TREES 
B.-M. Straub 
IPI, Institute for Photogrammetrie and Geolnformation, 30167 Hanover, Germany - bernd-m.straub@jipi.uni-hannover.de 
Commission IIT, WG I11/4 
KEY WORDS: Urban, Vegetation, Automation, Recognition, Algorithms, Texture, Infrared, High Resolution. 
ABSTRACT: 
In this paper we describe our recent work on the automatic extraction of trees from high resolution aerial images. In order to be more 
independent of color information we have investigated textural properties of trees and buildings. The aim is to be able to differentiate 
between object classes based on textural information. Texture is a characteristic feature of trees, and if color information is not 
available it is an important cue to differentiate between trees and buildings. The Gabor filter bank of the standardized MPEG-7 
Homogeneous Texture Descriptor (HTD) was used for the extraction of the textural properties. The qualification of the HTD for the 
extraction and classification of trees is evaluated. The evaluation is based on first experimental results, which are presented in the 
paper. 
1. INTRODUCTION 
Geographic information meets virtual reality (CROSSES, 
2002). The aim of the CROSSES (Crowd Simulation System 
for Emergency Situations) project is to develop a realistic 
training system for emergency forces. An important aspect of 
such a system is the use real data, which give the training staff a 
good impression of the local situation. One of the tasks that we 
had to solve in this project was the extraction of trees and 
buildings from aerial imagery. The production of the data for 
the 3D city model should be done automatically, because the 
system is proposed to be installed in different cities, always 
with actual real data. We have developed algorithms for the 
automatic extraction of buildings (Gerke et al., 2001) and trees 
(Straub and Heipke, 2001). For CROSSES colour infrared 
(CIR) aerial images were acquired in summer 2000. The image 
flight was carried out with 80% overlap along and across the 
flight direction. The image scale is 1:5000, which leads to a 
GSD of 10 cm at a scanning resolution 20um. Based on these 
images a digital surface model and a true orthoimage were 
automatically derived by the French company ISTAR (Gabet et 
al., 1994). The extraction of vegetation was planned from the 
beginning of the project, partly because reports had predicted an 
increasing request for vegetation in general, and trees in 
particular as a part of 3D city models (Fuchs et al, 1998). It 
seems, however, that CIR imagery is not readily available today, 
even though CIR imagery is well established for the extraction 
of vegetation information and does not handicap the extraction 
of man-made objects. Many customers of 3D city model data’ 
prefer true color images due to the appearance of the 
orthophotos. In order to be more independent of the available 
colors and especially of the infrared channel, we have decided 
to investigate textural properties for the extraction of trees 
during the vegetation period, and potentially their classification 
into different types. 
  
! We have learned that from discussions with other researchers 
and staff from companies working in the field of data 
production for 3D city models. 
There is no doubt that the use of the textural information is 
helpful for the detection of objects from images. Human 
analysts discriminate between areas with vegetation and trees 
and areas with man-made objects by using textural features 
(Haralick and Shapiro, 1992), and many promising results are 
reported in the literature regarding the use of texture for the 
automatic object extraction. But, there is no commonly accepted 
way to select the texture operators and to link the different 
textural features (Shao and Fórstner, 1994). 
Today, the description of texture is a part of the ISO/IEC 
standard MPEG-7, different texture descriptors were 
investigated by the MPEG consortium (MPEG-7, 2002). The 
Homogeneous Texture Descriptor (HTD) (Man Ro et al., 2001) 
which is composed of a Gabor filter bank, a formal description 
of the extracted features as well as different similarity measures, 
is investigated in this paper for the extraction of trees from high 
resolution aerial imagery. 
2. RELATED WORK 
The extraction of trees from optical and/or height data was 
investigated by different research groups. The discrimination of 
vegetation and man-made objects using true-color images is 
discussed in (Niederóst, 2000). Niederóst proposes the use of an 
artificial channel denoted as degree of artificiality, which can be 
computed from the red and the green band of true color images. 
(Brandtberg and Walter, 1998) have developed an approach for 
the extraction of trees from aerial images with a GSD of 10 cm 
based on the gray level curvature and length of edges in 
different scales. (Brunn and Weidner, 1997) proposed to use the 
variance of DSM surface normal to detect vegetation regions. 
Laser scanner data and a colour infrared image are used in 
combination by (Haala and Brenner, 1999) for the classification 
of an urban scene. A pixel based unsupervised classification 
algorithm is employed to perform the segmentation of the 
image. 
A - 351