stanbul 2004 
SE 
is composed 
domain with 
the spatial 
n modulated 
0+ ysen0)) 
filter. In the 
ussian curve 
of the Gabor 
yy @ 
T is centered 
in curve. 
the Gaussian 
(lated with 6 
50?) and 3 
on: F=0.3536 
0884 and c 
Iters covering 
lied and their 
y a Gaussian 
d for texture 
ides a robust 
scales. The 
n of a signal 
wavelets and 
ions of a base 
9 
G) 
anslation. The 
] by applying 
9 the original 
(6) 
filters, whilt 
alt of this, the 
'e into à series 
d fluctuations. 
image, and the 
cales or levels. 
International Archives of the Photogrammetry, Remote Sensing 
  
  
Since the most relevant texture information is lost in the 
lowpass filtering process, only fluctuations are used to calculate 
texture descriptors. If the inverse transform is applied to the 
fluctuations, three reconstructed images, or details, are 
obtained: horizontal, vertical and diagonal. This process is 
called multiresolution analysis. 
Regarding previous work in image texture analysis using 
wavelet decomposition, different texture features have been 
extracted, sometimes from the fluctuations and in other cases 
from the details, depending on the authors. Sometimes, basic 
features directly extracted from the histogram were used, such 
as the local energy (Randen and Husoy, 1999) or variance filter 
(Ferro and Warner, 2002). Simard et al. (1999), however, used 
wavelet histogram signatures, while Van de Wouwer et al. 
(1999) compared the energy, wavelet histogram signatures and 
coocurrence features. 
We compared the use of fluctuations and details, and four 
coocurrence features were calculated using them: variance, 
inverse difference moment, contrast and correlation. 
In a comparative study about the evaluation of the performance 
of texture segmentation algorithms based on wavelets, Fatemi- 
Ghomi et al. (1996) stated that the identification of the most 
appropriate parameters to use in a method is as important a 
decision as the choice of which method to use. We also wanted 
to know, given our particular classification cases, the best group 
of methodological parameters to solve each particular problem. 
The following parameters were tested: the type of features, the 
window or neighbourhood size, the type of wavelet, the 
influence of the level of decomposition, and the use of the sum 
of the details or the fluctuations, or to consider them 
independently. All these items will be analysed in the tests and 
results section. 
3. TESTS AND RESULTS 
The different texture analysis methods and parameters were 
evaluated for application in two environments: mediterranean 
forested areas and growing urban areas. In this section, we will 
first describe the testing areas and the type of image data used, 
then we will analyze the selection of the texture parameters. 
Finally, we will compare the accuracy of the classification 
obtained with the specific methods used, as well as the spectral 
Versus texture classification for one of the forest testing areas, 
where Quickbird images were available. 
3.1 Data and test areas 
Imagery from a total of four areas was used for evaluation, 
three forested and one urban, all in the mediterranean region of 
Spain. 
I. Forest 1: Located at the Sierra de Espadan, Castellon, 
near the central mediterranean coast of Spain, with 
dominance of forest (Pinus halepensis and Quercus 
suber) and shrubs (Quercus coccifera, Ulex,...), olive tree 
crops and rocky areas. Seven classes were defined: high- 
density forest, mid-density forest, areas combining forest- 
shrub, shrubs, scattered trees, scattered shrubs, and olive 
trees. For the purposes of evaluation, a mosaic image was 
Created from aerial orthophotos scanned to 1m of spatial 
resolution. 
Forest 2: This area is located slightly south and west of 
the previous one, in Ayora, Valencia, farther from the 
Coast and having a type of climate meso-mediterranean. 
  
1111 
and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
The vegetation of this area is mainly composed of forest 
(Pinus halepensis) and mediterranean shrub, usually 
mixed, and mountain crops (4migdalus communis, Olea 
europaea, Ceratonia siliqua) sometimes forming flat 
terraces on the sides of the mountains. The trees of this 
area are more scattered, in part because of a high 
recurrence of wildfires over the last several years. Nine 
classes were defined: high-density, mid-density and low- 
density forest, high-density and low density shrub, 
cereals, almond trees. reforestation areas, and crops on 
terraces. The data were digital orthophotos with Im of 
spatial resolution, that were also mosaicked to form an 
image with a variety of zones (figure 1). 
   
HD-forest 
TE LW 
3, 3 = 
A 
Pd A A 
ser 
SP 
  
  
    
Ne 
LD-forest HD-shrub 
NE DUNS 
EX 
I 
  
z 
Reforest. Terraces 
    
; Almond Cereal 
A z S ET 
Figure 1. Orthoimages mosaic of forest area 2, Ayora (left), and 
detail examples of eight of the classes defined (right). 
3. Forest 3: Located in the south of Menorca, one of the 
Balearic islands in the western Mediterranean sea. The 
landscape is composed of small forested areas (Pinus 
halepensis, Quercus ilex), and shrubs (Quercus coccifera, 
Ulex, Pistacia lentiscus, Rhamnus alaternus), usually 
combined with scattered trees (Olea europaea var. 
sylvestris), pasture areas, crops and residential areas. 
Seven forest and agricultural classes were defined: dense 
forest, forest-shrub, dense shrub, scattered trees, 
herbaceous vegetation or weeds, cereal or pasture, and 
Jallow; as well as two non-vegetation classes: residential 
areas and sea. In this case, a high-resolution panchromatic 
satellite image (QuickBird) was used, but resampled to 
2.4 m to keep visual coherence of the texture classes 
analysed, and to be able to compare them with the 
multispectral image from the same satellite. 
4. Urban: Located in the northern area of the city of 
Valencia, which, has experienced an important urban 
sprawl during the last several decades, and the 
surrounding towns. The classes considered were: old 
urban areas, new urban areas, more dispersed residential 
areas located outside of the city, industrial areas and 
barren soil, horticulture, and citrus fruit orchards. A 
panchromatic image captured by the satellite QuickBird 
was used, in this case resampled to 5 m. 
  
3.2 Selection of methodological parameters 
As we stated above, there are several methodological 
parameters that should be optimized for each type of 
application (forest or urban). We will now describe the results 
obtained in the parameter selection process, method by method. 
One of the most relevant parameters is the neighbourhood size, 
which is obviously related to the spatial resolution of the