Automated evaluation of generalized topographic maps

To meet the increasing demand for geospatial information at multiple scales, and to shorten update cycles for up-to-date geo-data, national mapping agencies (NMAs) and other data providers have introduced, or are considering the use of, (semi-)automated generalization systems in their production lines. Map generalization inevitably changes certain aspects (accuracy, completeness, etc.) of spatial data. Therefore, generalized data/maps should be evaluated to see whether they are in line with the map requirements defined for a target scale and for a certain task (e.g. visualization). The use of (semi-)automated generalization systems, on the other hand, increases the need for automated quality evaluation of generalized data/maps. Automated evaluation techniques provide more objective evaluation results and reduce the time needed for the subjective visual assessment. This thesis has identified the following issues from the state-of-the-art of automated evaluation of generalization output: • Lack of formalized map requirements: cartographic constraints were specified at a knowledge level that is only human readable, thus not suitable for automated evaluation. For example, complex concepts, relationships, and acceptable modifications in some constraints are not well defined. • Lack of balanced map requirements for a meaningful evaluation: previous research focused more on readability constraints and constraints for one object, or between two objects. The evaluation based on unbalanced constraints may lead to misleading (e.g. over-optimistic) results. • Lack of explicit information/knowledge, such as characteristics, relationships and patterns in topographic data, and links between corresponding entities between different scales. Abstract The main research question of this thesis is therefore “how to evaluate the quality of generalized general-purpose topographic data in a (semi)automatic way, so that a meaningful evaluation can be reached?” In the course of this research, I have made several contributions. To begin with, I have proposed an improved conceptual framework for automated evaluation of generalized maps, which consists of methodological and technological parts. The methodological part formulates empirical guidelines for selecting a set of evenly balanced constraints for a meaningful evaluation. To support automated evaluation of a set of evenly balanced constraints, e.g., readability, preservation, entity-level, property-/relationlevel constraints, and constraints at individual, group and feature class levels, I have proposed a three-step approach consisting of data enrichment, data matching and evaluation. In addition, this thesis has presented a first-order representation for multi-scale data that formalizes concepts of different granularities (i.e. objects, groups and feature classes) and complex relationships (e.g. semantic and contextual ones). Cartographic constraints can be precisely defined on top of this formal theory. Later in the design and implementation phase, I have focused on specific problems in automatically evaluating preservation constraints and the constraints on group level and feature class level. I have used building and road feature classes to demonstrate the proposed three-step evaluation approach, where settlement structures (e.g. building patterns) and spatial distributions of feature classes are extensively explored. This thesis contributes to the specific problems in the following ways: • I have proposed an improved approach to building pattern recognition in topographic data, where different pattern types recognized from different algorithms can be combined to obtain a better recognition; • I have proposed a generic matching approach based on pattern classification idea where various criteria can be aggregated to make decisions; besides, I have devised ad-hoc algorithms to match building patterns and alignments at different scales; • I have proposed an approach to evaluating building alignments in generalized data, making use of the results obtained in the above steps; • I have made attempts to measure spatial distributions of feature classes; two approaches are promising in comparing the similarity of spatial distributions: one is based on local densities and the other is based on distance. iiThe main research question of this thesis is therefore “how to evaluate the quality of generalized general-purpose topographic data in a (semi)automatic way, so that a meaningful evaluation can be reached?” In the course of this research, I have made several contributions. To begin with, I have proposed an improved conceptual framework for automated evaluation of generalized maps, which consists of methodological and technological parts. The methodological part formulates empirical guidelines for selecting a set of evenly balanced constraints for a meaningful evaluation. To support automated evaluation of a set of evenly balanced constraints, e.g., readability, preservation, entity-level, property-/relationlevel constraints, and constraints at individual, group and feature class levels, I have proposed a three-step approach consisting of data enrichment, data matching and evaluation. In addition, this thesis has presented a first-order representation for multi-scale data that formalizes concepts of different granularities (i.e. objects, groups and feature classes) and complex relationships (e.g. semantic and contextual ones). Cartographic constraints can be precisely defined on top of this formal theory. Later in the design and implementation phase, I have focused on specific problems in automatically evaluating preservation constraints and the constraints on group level and feature class level. I have used building and road feature classes to demonstrate the proposed three-step evaluation approach, where settlement structures (e.g. building patterns) and spatial distributions of feature classes are extensively explored. This thesis contributes to the specific problems in the following ways: • I have proposed an improved approach to building pattern recognition in topographic data, where different pattern types recognized from different algorithms can be combined to obtain a better recognition; • I have proposed a generic matching approach based on pattern classification idea where various criteria can be aggregated to make decisions; besides, I have devised ad-hoc algorithms to match building patterns and alignments at different scales; • I have proposed an approach to evaluating building alignments in generalized data, making use of the results obtained in the above steps; • I have made attempts to measure spatial distributions of feature classes; two approaches are promising in comparing the similarity of spatial distributions: one is based on local densities and the other is based on distance. ii To conclude, formalization of map specifications, data enrichment and data matching techniques form a promising framework and provide necessary instruments for automated evaluation of generalization output. This has been demonstrated by evaluating building alignments in generalized maps. Since this research does not provide a full solution to automated evaluation, a comprehensive evaluation integrating more constraints and feature classes for a whole map is a topic for further research. This would refine and improve the current empirical guidelines for selecting a set of evenly balanced constraints for a meaningful evaluation. In addition, this research has discussed potential drawbacks of the methods proposed to address the above-mentioned specific problems and outlined ways in which they can be improved. Also this requires further research.