ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision“, Graz, 2002 
A COMBINED ESTIMATION-DEFORMATION MODEL FOR AREA DETECTION: 
APPLICATION TO TOPOGRAPHIC AREA FEATURE UPDATE 
Sylvie Jodouin, Layachi Bentabet, Djemel Ziou, Jean Vaillancourt, Costas Armenakis 
Université de Sherbrooke, Québec, Canada 
KEY WORDS: active contour, multispectral boundary finding, region-based segmentation, MAP, change detection, topographic 
databases update, Raster and vector information, integration. 
ABSTRACT: 
This paper presents a fully automated approach for area detection based on multi-spectral images and features from a topographic 
database. The vectors residing in the database are refined using active contours (snakes) according to updated information provided 
by the multi-spectral images. The conventional methods of defining the external energy of the snake based on statistical measures 
or gradient-based boundary finding are often corrupted by poor image quality. Here a method to integrate the two approaches is 
proposed using an estimation of the maximum a posteriori (MAP) segmentation in an effort to form a unified approach that is 
robust to noise and poor edges. We further propose to improve the accuracy of the resulting boundary location an update of the 
snake topology. A number of experiments are performed on both synthetic and LANDSAT 7 images to evaluate the approach. 
1. INTRODUCTION 
Topographic databases are gaining popularity as a reference 
tool in many fields of application. The providers of topographic 
information are currently concerned with how to maintain data 
updated and also how to increase its accuracy with limited 
resources. Moreover, the evolving needs for current basic 
spatial data requires reliable and fast processing methods to 
address these concerns. This problem could be overcome 
through image processing techniques, which allow increasing 
accuracy of both geometric and temporal features. An 
increasing number of methods for updating spatial information 
based on image processing began to appear in the latest years. 
From the proposed methods, we are mostly interested with 
those based on the snakes. The proposed approaches deal with 
different type of images, mainly with single band satellite 
(Bentabet et al, 2001; Horritt, 1999) or airborne (Auclair- 
Fortier et al., 2001) images and from different type of sensors 
such as radar or optical. The snake model is defined according 
to the geometry of the target feature. Some works present 
linear snakes to search linear features such as road (Bentabet 
et al, 2001; Auclair-Fortier et al., 2001). Others present closed 
snakes to obtain a description of area features such as water 
regions (Horritt, 1999). 
In this paper, we propose a method to update existing area 
features from a given topographic database using of multi- 
spectral images. The available database vectors provide an 
interesting initialization for the area localization process. In 
this context, the closed snake approach was presented as a 
natural solution. The main contribution of this work is the 
formulation of the external forces, which deform the snake by 
combining both statistical and boundaries information. 
The region-based and the boundary finding measures provide 
us with complementary information. By reviewing the existing 
works focusing on the integration of region-based information 
with boundary information, we conclude that they are mainly 
made within image segmentation framework (Chakraborty and 
Duncan, 1999). The boundary-based methods have superior 
localization properties. Also, they are robust to changes in the 
gray-level distribution since they look at the derivative 
information. However, they often give high rate of false edges 
due to textured regions. In addition, a weak response is 
produced when far from the boundaries of the area. The region- 
based and especially statistical methods supply a good model 
for textured regions and good response when far from the 
edges. Furthermore, region-based methods have the advantage 
of being less susceptible to noise than other methods that 
involves derivative information. Unfortunately, these methods 
suffer from the problems of poor localization and over- 
segmentation. 
As pointed out in the above discussion, the region and 
boundary based methods have their different advantages and 
disadvantages. This brings us to the fact that integration 
methods are likely to perform better than either of the methods 
alone. An integration method will combine the complementary 
strength of these individual methods and decrease their 
drawbacks, as pointed out in (Chakraborty and Duncan, 1999; 
Pavlidis and Liow, 1990; Tek and Kimia, 1995). Many works 
have addressed the problem of combination of region-based 
and boundary-based methods. Some studies focus on the Al 
techniques, which define a set of rules in order to deal with 
conflicting situations (Pavlidis and Liow, 1990). Another way 
of achieving combination is the reaction-diffusion method (Tek 
and Kima, 1995). However, the problem is that if any one of 
the processes makes error (e.g., a false edge), it is propagated 
to the final solution. Chakraborty et al. propose to use the game 
A - 181