Tal Abramovich 
  
presented here is based on a fuzzy logic concept. The use of fuzzy logic is useful for structuring the knowledge 
about geometrical, radiometrical and contextual properties of roads at different sources. It enables the 
representation of non determined properties, such as “width,” as a fuzzy input, rather than a strict “wide” or 
“narrow” values. 
Examples for the use of the fuzzy theory in the context of road extraction can be found in Agouris at el. (1998). 
In that work, the algorithms are designed to function within an integrated geospatial environment. Fuzzy logic is 
used in the first stage for detecting road pixels, i.e., pixels that have high probability to belong to a road. In a 
second stage, road tracking is performed, based on the selected pixels from the first stage, and some geometric 
properties. 
In the work described in this paper, the idea of using fuzzy logic for matching road information from GIS to 
features extracted from an aerial image is presented. The paper describes the matching problem, details the 
proposed approach and shows some preliminary results. 
2 MATCHING GIS ROAD ENTITIES WITH IMAGE ROAD SEGMENTS 
Given two sets of features, one from a GIS (reference data) and one extracted from a digital image (extracted 
data), the matching task in this work is to conjugate as many road features from the two sets as possible. A road 
entity in the reference data may have a match in the extracted data, or may not have one. The matching step 
should generate these pairs with a minimum number of errors. 
In order not to miss important information, it is possible that at the initial state a road segment in the reference 
data will have several potential matches in the extracted data. Avoiding combinatorial explosion is possible by 
limiting the search space to a predefined buffer along the road segment. The final purpose of the matching is to 
find the hypothesized road segment that has a best fit to the reference segment. However, once one of the 
features is selected as “best,” the question that remains is “is this the correct match?” 
The matching process might entail two types of errors in the context of roads. The first is a false negative result, 
meaning that entity in the reference data was not matched to a hypothesized road in the extracted data, while the 
latter is in fact a road. The second type of error occurs if an entity in the reference data is matched to a 
hypothesized road, when the latter is not a road. This type of error is denoted as false positive. The goal of the 
method described in this paper is to reduce the possibility of making such errors. The idea is to analyze the road 
model and structure the matching process in a way that the similarity between two features extracted from 
different sources is quantified. This provides a better estimation of the success of the matching process. 
Quantifying the similarity is based on a comparison of geometric (and possibly other) properties, and by 
weighting these properties according to a prior knowledge about roads. 
In the first stage, a generic road model is defined which is mainly linked to the geometric and radiometric 
properties of the road. For example, (i) roads have parallel edges; (ii) there are no sharp radiometric changes 
along the road; (iii) a road profile changes gradually; (iv) road width is constant, or changes gradually; (v) roads 
have a maximum curvature; and (vi) road height varies in a rather gradual manner. Baumgartner af el. (1997), 
for example, took the two first fundamental assumptions for their road model. Each property is assigned a 
weight value that denotes its significance. A major idea in this research is that both hypothesized roads in the 
image and road entities from the GIS share some common attributes. Therefore, the matching tool is based in 
part on comparing these attributes from the two sources. 
The entire matching process is formulated as a fuzzy problem. A description of the idea is given in the following 
section. 
3 THE PROPOSED ALGORITHM 
The proposed algorithm for quantifying the similarity between a road segment from a GIS and a hypothesized 
road segment extracted from an image is based on fuzzy logic. Zadeh (1973) conceived the concept of fuzzy 
logic as a way of processing data by allowing a partial set membership rather than a crisp set membership or 
non-membership. Fuzzy logic provides a simple way to achieve a quantitative conclusion based upon vague, 
ambiguous, imprecise, noisy or incomplete input information. 
  
18 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
The 
outp 
term 
the c 
Each 
usua 
a val 
“the 
The 
seg 
show 
Fig 
The c 
The r 
ss 
Table 
know 
it can 
shouk