The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B3b. Beijing 2008 
et al., 2002) extract buildings from Digital Elevation 
Model(DEM); Huber et al(Huber et al.) use LIDAR (Light 
Detection And Ranging) as an attractive alternative for the 
building extraction and acquisition of 3D information. 
The object-oriented building extraction methods have 
developed quickly, and a lot of work has been done. Muller 
(Muller et al, 2005) presented a building extraction approach 
that divides the method into a low-level image processing step 
and a high-level feature extraction and final classification step. 
They use seeded-region growing algorithm to segment the 
entire image, then take area, mean hue angle as preselection 
features to reduce calculation time, finally classify based on 
robust features as size, form, hue angle, neighbourhoods, and 
shadow. Pasko and Gruber (1996) gave an approach that 
incorporates a priori knowledge for buildings extraction. In 
Jiang(2004) multiple high-resolution images and Digital 
Surface Model(DSM)are combined to extract the urban road 
grid in complex, stereotypical, residential areas. In our 
approach, we proposed an object-oriented building extraction 
method by DSM and orthoimage. In Section2, the main steps of 
building extraction are given. Section 3 illustrates an example 
for building extraction and precision evaluation. Conclusions 
are given in Section4. 
2.1 Image Segmentation 
Detection, segmentation and classification of specific objects 
are the key building extraction blocks of object-oriented 
approach in image analysis area. Image segmentation is the first 
step and also one of the most critical tasks of image analysis 
(Zhang, 2006). In image processing, segmentation is the 
subdivision of a digital image into smaller partitions according 
to given criteria. The fundamental step of image analysis is a 
segmentation of an image—into image object primitives. 
Segmentation algorithms are used to subdivide the entire image 
represented by the pixel level domain or specific image objects 
from other domains into smaller image objects. For 
segmentation, the basic algorithms like chessboard, quadtree, or 
multi-resolution segmentation usually are used. We emphasize 
multi-resolution segmentation. It is a bottom-up segmentation 
algorithm based on a pairwise region merging technique, and it 
is an optimization procedure which, for a given number of 
image objects, minimizes the average heterogeneity and 
maximizes their respective homogeneity. 
The segmentation procedure works according the following 
rules, representing a mutual-best-fitting approach (Definiens, 
2007): 
2. OBJCET-ORIENTED BUILDING EXTRACTION 
Pixel-oriented processing approaches deal with each pixel, so 
when the resolution improved, the number of pixels consisted in 
the building increased sharply. Object extraction is time 
consuming. Also, Salt and Pepper Noise phenomenon in 
classification and extraction is evident and not easy to get rid of. 
Of all the disadvantages, the utmost one is it has excessive 
dependence on the spectral information. Object-oriented 
method can solve these problems well. It can take the full 
advantages of high-resolution of remote sensing images, 
integrate the semantic, contextual, and spectral information, and 
mine the hidden information sufficiently. At the same time, it 
deals with objects that are consisted of pixels with the same 
grey level. It reduces time consuming of computation. What’s 
more object-oriented methods work more in line with people's 
cognitive thinking habits and characteristics. In this paper, 
object-oriented approach is selected to extract building 
information from urban areas. 
The object-oriented building extraction from high-resolution 
images typically includes several steps: data pre-processing, 
multi-resolution image segmentation, the definition of the 
characteristics used to delineate the buildings, building 
extraction, post editing and accuracy evaluation. The flow chart 
of the paper is shown as below: 
Figure 1. The flow chart of building extraction 
A. The segmentation procedure starts with single image objects 
of 1 (one) pixel size and merges them in several loops 
iteratively in pairs to larger units as long as an upper threshold 
of homogeneity is not exceeded locally. This homogeneity 
criterion is defined as a combination of spectral homogeneity 
and shape homogeneity. Scale parameters can be modified 
manually through calculation. Higher values for the scale 
parameter result in larger image objects, smaller values in 
smaller image objects. 
B. As the first step of the procedure, the seed looks for its best 
fitting neighbor for a potential merger. 
C. If best-fitting is not mutual; the best candidate image object 
becomes the new seed image object and finds its best fitting 
partner. 
D. When best fitting is mutual, image objects are merged. 
E. In each loop, every image object in the image object level 
will be handled once. 
F. The loops continue until no further merger is possible. 
The homogeneity criterion is calculated as a combination of 
color and shape properties of both the initial and the resulting 
image objects of the intended merging. Here the color 
homogeneity is based on the standard deviation of the spectral 
colors. The shape homogeneity is based on the deviation of a 
compact or a smooth shape. 
Homogeneity 
Criterion a 
Color 
Shape 
(t-w,l jjrj] Smoothness 
M Compactness 
Set weightings pâiâtwlen 
in the WR Process dialog box, 
Figure 2. Weighted components of the homogeneity criterion 
(Definiens, 2007) 
The new image objects created by segmentation are stored in 
what is called a new image object level. Each image object is 
defined by a contiguous set of pixels, where each pixel belongs