Full text: Technical Commission III (B3)

   
  
  
  
  
  
  
   
  
   
   
  
  
  
   
    
    
   
   
    
   
   
    
    
   
    
  
  
    
    
    
   
    
   
   
   
    
   
    
   
   
FEATURE MODELLING OF HIGH RESOLUTION REMOTE SENSING IMAGES 
CONSIDERING SPATIAL AUTOCORRELATION 
    
   
Y. X. Chen*, K. Qin* *, Y. Liu, S. Z. Gan?, Y. Zhan*? 
* School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, China 
? Information Technology Simulation Teaching and Research Section, Institute of Chemical Defense of CPLA 
Beijing, China 
Commission III, ICWG III/VII 
KEY WORDS: high resolution, feature, modelling, spatial autocorrelation, segmentation 
ABSTRACT: 
To deal with the problem of spectral variability in high resolution satellite images, this paper focuses on the analysis and modelling 
of spatial autocorrelation feature. The semivariograms are used to model spatial variability of typical object classes while Getis 
statistic is used for the analysis of local spatial autocorrelation within the neighbourhood window determined by the range 
information of the semivariograms. Two segmentation experiments are conducted via the Fuzzy C-Means (FCM) algorithm which 
incorporates both spatial autocorrelation features and spectral features, and the experimental results show that spatial autocorrelation 
features can effectively improve the segmentation quality of high resolution satellite images. 
1. INTRODUCTION 
High spatial resolution remote sensing imagery obtained from 
satellite (IKNOS, Quickbird, GeoEye-1, WorldView-2, etc) and 
airborne sensors have become increasingly available in recent 
years (Johnson & Xie, 2011). These data provide amazing 
details of the Earth's surface, but for information extraction 
from complex scene such as urban environment, it is difficult to 
obtain satisfactory results using only spectral information (Byun 
et al., 2011). 
It is well known that combining spatial and spectral information 
is a good strategy to improve urban land use classification. 
Features extracted by using co-occurrence matrices, Gabor 
wavelets, morphological profiles, and Markov random fields 
have been widely used in the literature to model spatial 
information in neighborhoods of pixels (Akcay & Aksoy, 2008). 
Spatial autocorrelation as spatial information is an inherent 
feature of remote sensing data and a reliable indicator of 
statistical separability between spatial objects. In remote 
sensing, spatial autocorrelation means the spectral dependence 
existing between a pixel and its neighbors, that is, spectral value 
of a pixel is usually not independent but correlated with those of 
its neighboring ones. Spatial autocorrelation provides us the 
structural information between spectral values of pixels, which 
is usually more stable and robust to noise than individual pixel. 
This information may be used to improve the segmentation 
quality or classification accuracy for spectrally heterogeneous 
classes and overcome the current spectral limitations of very 
high spatial resolution satellite images. 
The basic approach modelling spatial autocorrelation is to use 
spatial autocorrelation statistics, including global statistics and 
local statistics. Global statistics of spatial autocorrelation such 
as Moran's / and Geary's C, are simple summary measures 
which are difficult to uncover the local spatial variability. Getis 
  
* Corresponding author: E-mail address: qink(g)whu.edu.cn 
   
statistic (Ord & Getis, 1995) is a measure of local spatial 
autocorrelation, which is quite effective in distinguishing “hot 
spots” and "cold spots". Thus, it could be used, for example, to 
identify a group of bright or dark pixels that represent a spectral 
response from a homogeneous feature (Myint et al., 2007). 
Another approach modelling spatial autocorrelation is 
semivariogram, which is a geostatistical function and can be 
used to model spatial variation patterns of typical object classes 
in the image, providing structure information of spatial 
autocorrelation. The range of the semivariogram can be used as 
a measure of spatial dependency or homogeneity (Franklin et al., 
1996) and it has been proved to be directly related to the size of 
objects or patterns in an image (Balaguer et al, 2010). 
Therefore, it may be used to determine the proper window size 
for each pixel in local spatial autocorrelation analysis. 
This paper focuses on the analysis and modelling of spatial 
autocorrelation features for improving the segmentation quality 
of high resolution satellite images. The semivariograms are used 
to model spatial variability of typical object classes, while Getis 
statistic is used to calculate the local spatial autocorrelation 
based on range information provided by semivariograms. Two 
segmentation experiments based on Fuzzy C-Means (FCM) 
clustering algorithm (Bezdek, 1981) are conducted. The results 
show that spatial autocorrelation features can effectively 
improve the segmentation quality of high resolution satellite 
images. 
2. STUDY AREA AND DATA 
In this paper, the experimental data are Quickbird images of two 
different sites in Wuhan, China, with the resolution of 
panchromatic band 0.61 m and multi-spectral band 2.44 m. The 
image sizes of the two sites are 798 pixels X 642 pixels and 349 
  
  
pixels X 22 
bands (ban 
The color i 
and by fiel 
vegetation, 
object clas 
lands. Mos 
in the imag 
Fig 
31 Sem 
The semiv 
the spatiz 
experimen 
Y | 
where z (x 
the variab] 
x; to pixel 
+h. 
Semivario 
sill and ra 
(or lag) 
variability
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.