In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). IAPRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
different error types based on FMV. thus minimizing the 
misclassification errors. After describing the study areas and 
data sources in the following section, this paper is organised as 
follows. Section 4 describes the methods. Section 5 presents 
and evaluates the results and we summarise our results in 
Section 6. 
3. STUDY AREAS AND DATA SOURCES 
3.1 Test Zones and Input Data 
Four test datasets of different sensor and scene characteristics 
were used in this study as summarized in Table 1 and 2. Test 
area 1 is a part of the region surrounding the University of New 
South Wales campus, Sydney Australia, which is a largely 
urban area that contains residential buildings, large Campus 
buildings, a network of main roads as well as minor roads, 
trees, open areas and green areas. The colour imagery was 
captured by film camera at a scale of 1:6000. The film was 
scanned in three colour bands (red, green and blue) in TIFF 
format, with 15pm pixel size (GSD of 0.09m) and radiometric 
resolution of 16-bit as shown in figure 1(<7). Test area 2 is a pan 
of Bathurst city, NSW Australia, which is a largely rural area 
that contains small residential buildings, road networks, trees 
and green areas. The colour (red, green and blue) images were 
captured by a Leica ADS40 line scanner sensor and supplied as 
an ortho image as shown in figure 1(Z>). Test area 3 is over 
suburban Fairfield, NSW Australia covering low density 
development in the southwest half of the scene, and large 
industrial buildings in the northeast part as shown in figure 1(c). 
The image data was acquired by a film camera at a scale of 1:10 
000 which was scan digitized and supplied as an ortho image. 
Test area 4 is over Memmingen Germany, featuring a densely 
developed historic centre in the north of the scene and industrial 
areas in the remainder as shown in figure ](d). Multispectral 
images (CIR), including an infrared image with the same 
resolution as the colour bands, were acquired by a line scanner 
sensor and supplied as an ortho image. 
Figure 1. Orthophotos for: (a) UNSW; (b) Bathurst; (c) 
Fairfield; and (d) Memmingen. 
Test area 
Size 
(Km) 
bands 
pixel 
size 
(cm) 
Camera 
UNSW 
0.5 X 0.5 
RGB 
9 
LMK1000 
Bathurst 
1 X 1 
RGB 
50 
ADS40 
Line scanner 
Fairfield 
2x2 
RGB 
15 
LMK1000 
Memmingen 
2x2 
CIR 
50 
TopoSys Falcon 11 
line scanner 
Table 1. Characteristics of image datasets. 
UNSW 
Bathurst 
Fairfield 
Memmingen 
Optech 
ALTM 
1225 
Leica 
ALS50 
Optech 
ALTM 
3025 
TopoSys 
Spacing across 
1.15 
0.85 
1.2 
0.15 
track (m) 
Spacing along 
1.15 
1.48 
1.2 
1.5 
track (m) 
Vertical 
0.10 
0.10 
0.15 
0.15 
accuracy (in) 
Horizontal 
0.5 
0.5 
0.5 
0.5 
accuracy (m) 
Density 
1 
2.5 
1 
4 
(Points/m 2 ) 
Sampling 
11 
150 
167 
125 
intensity 
(mHz) 
Wavelength 
1.047 
1.064 
1.047 
1.56 
(pm) 
Laser swath 
800 
777.5 
700 
750 
width (m) 
Recorded 
1*' and 
I я and 
I s ' and last 
1 sl and 
pulse 
last 
last 
last 
Table 2. Characteristics of lidar datasets. 
3.2 Training Datasets 
All tests were conducted using identical training sets. Eighty 
polygons of approximately equal areas, twenty for each land 
cover class, buildings, trees, roads and ground, were overlaid 
over each image to generate the training data. The positions of 
the polygons were selected carefully to be representative and to 
capture changes in the spectral variability of each class. The 
training data for each test area consists of 1644. 1264, 1395 and 
1305 training pixels for buildings, trees, roads and ground 
respectively for each band of the input data. Class “ground” 
mainly corresponds to grass, parking lots and bare fields. 
3.3 Reference Data 
In order to evaluate the accuracy of the results, reference data 
were captured by digitising buildings, trees, roads and ground 
in the orthophotos. Class “ground” mainly corresponds to grass, 
parking lots and bare fields. We chose to digitize all 
recognisable features independently of their size. Adjacent 
buildings that were joined but obviously separated were 
digitized as individual buildings. Otherwise, they were merged 
as one polygon. In order to overcome the horizontal layover 
problem of tall objects such as buildings, roofs were first 
digitized and then each roof polygon was shifted if possible so 
that at least one point of the polygon coincided with the 
corresponding point on the ground. For Fairfield, the 
orthophoto and the lidar data correspond to different dates. 
Thus, we excluded from the analysis 41 building polygons that 
were only available in one data set. Larger areas covered by 
trees were digitised as one polygon. Information on single trees 
was captured where possible.