CONTEXT-BASED CLASSIFICATION OF OBJECTS IN CARTOGRAPHIC DATA

Cartographic Journal, …, 2008

Matching of building polygons with different levels of detail is crucial in the maintenance and quality assessment of multirepresentation databases. Two general problems need to be addressed in the matching process: (1) Which criteria are suitable? (2) How to effectively combine different criteria to make decisions? This paper mainly focuses on the second issue and views data matching as a supervised pattern classification. Several classifiers (i.e. decision trees, Naive Bayes and support vector machines) are evaluated for the matching task. Four criteria (i.e. position, size, shape and orientation) are used to extract information for these classifiers. Evidence shows that these classifiers outperformed the weighted average approach.

4 th AGILE Conference on Geographic …, 2001

In this paper we describe the application of analogical structure matching techniques to the domain of Geographic Information Systems (GIS). Automatic categorisation of map data into roads, buildings etc. is currently based on isolated objects. We describe how identifying analogous clusters of objects can categorise ambiguous polygons by introducing context into the categorisation process. We describe a number of GIS classification tasks that can be performed by analogical structure matching, resulting in a useful classification tool that operates in a cognitively plausible manner.

2008

In this paper the conversion of a spaghetti into a topologically structured topographic base map is discussed. The first step, called structuring, comprises node computation, handling overshoots, undershoots and overlapping line segments. Node computation is a computational geometry problem driven by tolerances and a weighting scheme. The next step is placing additional lines by hand to close certain areas. Finally, area features are classified using a rule-based system. Specifically the rule-based classification of the areas has been implemented, tested and compared to human classification. In these experiments automatic classification leads to a speed-up of a factor 2 while maintaining a similar classification performance when compared with manual classification. 1

In this study, car navigation is considered as the basic activity and the multiple representation approach for the production of the navigation maps is taken into account to derive road network data for different representation levels. In this context, road junctions, which are the most complicated parts of road networks, are classified considering two different approaches, an approach based on cartographic aspects and an approach based on algorithmic aspects. In the first approach, road junctions are classified according to their aspects at different representation levels. In this concept, three representation levels, which are derived from 1:5000 scaled base level, are used and representations of possible road junctions are structured according to characteristics of these levels. Finally, they are classified by considering their geometrical shapes in each level. In the second approach, road junctions are classified according to algorithmic concerns. Junctions are defined in matric...

ISPRS International Journal of Geo-Information

A methodology for matching bidimensional entities is presented in this paper. The matching is proposed for both area and point features extracted from geographical databases. The procedure used to obtain homologous entities is achieved in a two-step process: The first matching, polygon to polygon matching (inter-element matching), is obtained by means of a genetic algorithm that allows the classifying of area features from two geographical databases. After this, we apply a point to point matching (intra-element matching) based on the comparison of changes in their turning functions. This study shows that genetic algorithms are suitable for matching polygon features even if these features are quite different. Our results show up to 40% of matched polygons with differences in geometrical attributes. With regards to point matching, the vertex from homologous polygons, the function and threshold values proposed in this paper show a useful method for obtaining precise vertex matching.

Artificial Intelligence in Design ’00, 2000

We introduce a shape categorisation method for architectural drawings that takes shape characteristics derived from those drawings as features. We propose a feature-based shape analysis procedure and a formal model of shape categorisation based upon the measurement of featural similarity of shapes. Shape categorisation is demonstrated in experiments on architectural drawings. 1 Q-codes have been developed as a qualitative shape encoding scheme, which converts shapes into systematically constructed qualitative symbols. Four separate Q-codes-A-, L-, K-and C-code-encapsulates shape characteristics in terms of angle, relative length, curvature and convexity (Park, 1999).

International Conference of Young Professionals «GeoTerrace-2020», 2020

Integration of geospatial datasets is complex and time-consuming process. Integration is especially difficult if the geospatial datasets do not share common feature identification attributes. In this study, we explored the software implementation of the method of spatial identification polygonal features of geospatial datasets by detecting matching pairs of features from two different datasets using measures of their spatial overlap and the similarity of their morphometric characteristics (perimeter, area, widthto-length ratio, blockiness, number of vertices). The polygonal feature identification program is implemented in PL/pgSQL for a spatial database in PostgreSQL/PostGIS. The results of the experiment on the example of the 13832 polygonal building models for the town of Bila Tserkva using an OSM dataset and a digital topographic plan of 1:2000 scale showed the ability to perform spatial identification of about 85% of buildings with little or no user intervention.

2006

In the GIS domain we are often faced with a great amount of shape-related data. Therefore, it is a challenging task to find concise description approaches which support the efficient retrieval of specific objects. In order to address this demand we apply a method that has recently been introduced in the context of shape-based image retrieval of two-dimensional silhouettes, namely the scope histogram. Scope histograms pertain to the group of qualitative shape descriptions as they characterise a shape by the general configuration of its parts. In particular, scope histograms allow the comparison of two shapes with constant time complexity. Despite of its low complexity, the approach achieves promising retrieval results. However, up to now the definition of scope histograms is limited to closed polygons. In this paper we investigate the application of scope histograms to the GIS domain. Since the contour of silhouettes is always closed, a restriction to closed polygons is no limitation in that domain. By contrast, it frequently is when dealing with GIS data. In this domain, we are rather often faced with open polygons; think for example of courses of rivers, borders, and coastlines. Therefore, we modify the original definition of scope histograms in order to be able to handle arbitrary polygons. Although our new definition leads to a more compact description than the original one, retrieval results are even improved by this modification.

Landscape and Urban Planning, 2012

This paper presents a technique for land-use object-based image classification of urban environments that combines high spatial resolution multi-spectral imagery and LiDAR data. Cadastral or land registry plots are used to divide the image and define objects. A set of descriptive features is presented, describing the objects at different urban aggregation levels. Objects are characterised by means of image-based (spectral and textural), three-dimensional, and geometrical features. In addition, contextual features describing two levels of the object are defined: internal and external. Internal contextual features describe the land cover object types (buildings and vegetation) inside the object. External contextual features describe each object while considering the common properties of neighbouring objects, which in urban areas usually coincide with the urban block. The proposed descriptive features emulate human cognition by numerically quantifying the properties of the image elements that enable their discrimination. The land-use classification accuracy values show that the proposed descriptive features enable an efficient characterisation of urban environments. The complementariness between the features derived from different aggregation levels is noticeable. Image-based features are highly discriminative, and the addition of internal and external contextual features significantly increases the classification accuracy of the urban classes considered in this study.