nbul 2004 
oin, ES. 
of radar 
S, 30, pp. 
ve remote 
S, 34, pp. 
fications 
-518. 
M. Haken 
of conifer 
& P de 
nperature 
p.9-13. 
systems. 
isi, 2001. 
diometer 
513. 
ives with 
northern 
pp. 388- 
scattering 
! Dp. 177- 
CONTEXTUAL CLASSIFICATION OF 
REMOTELY SENSED DATA USING MAP APPROACH AND MRF 
R. Khedam", A. Belhadj-Aissa" 
* Image Processing Laboratory, Electronic and Computer Science Faculty, Technology and Sciences University 
(USTHB), B.P. 32, El Alia, Bab Ezzouar, 1611, Algiers, Algeria. 
radjakhedam(@lycos.com, h.belhadj@lycos.com 
Commission PS WG VII/1 
KEY WORDS: Remote sensing, classification, supervised, Bayes, contextual, Markov, optimisation. 
ABSTRACT: 
Classification of land cover is one of the most important tasks and one of the primary objectives in the analysis of remotely sensed data. 
Recall that the aim of the classification process is to assign each pixel from the analysed scene to a particular class of interest, such as 
urban area, forest, water, roads, etc. The image resulting from the labelling of all pixels is henceforth referred to as “a thematic map”. 
Such maps are very useful in many remote sensing applications especially those concerned with agricultural production monitoring, land 
change cover and environmental protection. Conventional classification methods commonly named “punctual methods", classify each 
pixel independently by considering only its observed intensity vector. The result of such methods has often *a salt and pepper 
appearance" which is a main characteristic of misclassification. In particular of remotely sensed satellite imagery, adjacent pixels are 
related or correlated, both because imaging sensors acquire significant portions of energy from adjacent pixels and because ground cover 
types generally occur over a region that is large compared with the size of a pixel. It seems clear that information from neighbouring 
pixels should increase the discrimination capabilities of the pixeFbased measured data, and thus, improve the classification accuracy and 
the interpretation efficiency. This information is referred to as the spatial contextual information. In recent years, many researchers have 
proven that the best methodological framework which allows integrating spatial contextual information in images classification is Markov 
Random Fields (MRF). In this paper, we shall present a contextual classification method based on a maximum a posterior (MAP) 
approach and MRF. An optimisation problem arises and it will solved by using an optimisation algorithm such as Iterated Conditional 
Modes (ICM) which occurs the definition and the control of some critical parameters : neighbouring size, regularisation parameter value 
and criterion convergence. Test data available is SPOT image of “Blida” region sited at 50km on the south west of Algiers (Algeria). 
This image acquired on February 1986, contains seven main classes. The result of our contextual classification process is an interpretable 
and more easily exploitable thematic map. 
or global scales. Remote sensing is a collective name for several 
techniques which study at distance the ground surface or the 
1. INTRODUCTION 
In the current decade, global environmental change has reached atmosphere. Sensors installed on satellites or airplanes receive 
beyond the research domain and become a major national and and/or send radiation to the earth. The variation in amount and 
international policy issue. The project "Analysis of multitemporal wavelength of the reflected energy between studied objects or 
remotely sensed data ; multispectral and interferometric SAR phenomena gives the object its spectral signature and makes it 
imagery for land cover change in northern Algeria" was possible to distinguish between different types of land use, 
established in our laboratory on January 2004. The project has the vegetation, soils etc. Remotely sensed data are being and will 
objective of analysing the spatial characteristics, temporal continue to be used to retrieve information on a land cover map 
dynamics, and environmental consequences of land-use and land- which hold an important place at each step of a territory planning 
cover changes which have occurred in northern Algeria over the project. For a better characterization of land cover mapping, data 
period of 1980 and 2004 as a result of a range of socio-economic, classification approaches are generally proposed to obtain also a 
biogeophysical and natural driving forces. Especially, over the last robust objects or classes identification. Conventional automatic 
three years, northern Algeria has known two natural catastrophes classification techniques called also “punctual classifications” 
; flood and mudflows of Bab El Oued city happened on classify each pixel independently without tacking into account 
November 10, 2001 and a strong earthquake which struck information given by its context which a very helpful information 
Boumerdes city on May 21, 2003. These two events have caused because the response and class of two or more spatially 
land cover changes, land degradation and serious materials neighboring pixels are highly related. Different approaches have 
damage. The data analysis is used to project plausible future been taken to incorporate context in classification of remote 
changes in land use and land cover under different assumptions of sensing data and have named "contextual classification" 
future natural, demographic, economic, technological, social and (Kartikeyan ef al., 1994). We find approaches based on clustering 
political development. Given the current techniques available, pixels of the image according to the similarity of their response 
remote sensing is recognized as an efficient tool for earth (Amadamn et al., 1988, Kettig et al., 1976), relaxation techniques 
watching and land monitoring and provides the most feasible where probabilities of neighboring pixels are used iteratively to 
approach to land surface change detection at regional, continental update the probability of a given pixel (Richards ef a/., 1981) and 
11