Laura Keyes

Using moment invariants for classifying shapes on large-scale maps

Automated feature extraction and object recognition are large research areas in the field of image processing and computer vision. Recognition is largely based on the matching of descriptions of shapes. Numerous shape description techniques have been developed, such as scalar features (dimension, area, number of corners etc.), Fourier descriptors and moment invariants. These techniques numerically describe shapes independent of translation, scale and rotation and can be easily applied to topographical data. The applicability of the moment invariants technique to classify objects on large-scale maps is described. From the test data used, moments are fairly reliable at distinguishing certain classes of topographic object. However, their effectiveness will increase when fused with the results of other techniques.

Comparing learning strategies for topographic object classification

Two methods of topographic object classification through shape are described. Unsupervised classification through clustering analysis is compared with supervised classification based on a Bayesian framework. Both are applied to the real world problem of checking and assigning feature-codes in large-scale topographic data for use in computer cartography and Geographical Information Systems (GIS). Categorisation is accompanied by a confidence measure that the classification is correct. Both types of classification were implemented and their outcomes evaluated and compared. As a case study, results and conclusions are presented on the classification and identification of archaeological feature shapes on OS large-scale maps. It was found that the supervised classification model used out-performed the unsupervised classification model to a considerable degree.

Shape Description for Automatically Structuring Graphical Data

This work explores automatic object recognition and semantic capture in vector graphics through shape description. The low-level graphical content of graphical documents, such as a map or architectural drawing, are often captured manually and the encoding of the semantic content seen as an extension of this. The large quantity of new and archived graphical data available on paper makes automatic structuring of such graphical data desirable. Contour shape description techniques, such as Fourier descriptors, moment invariants play an important role in systems for object recognition and representation. However, most work carried out in this area has concentrated on categories of object boundaries representing very specific shapes (for example, a particular type of aircraft). Two classifiers were implemented and proved accurate in their automatic recognition of objects from drawings in different domains. Classical classifier combination techniques were used to improve performance. Further work will employ more complex fusion techniques and it is envisaged they will be used in combination with recognition based on object context using various modelling methods. A demonstration system has been constructed using all these techniques.

Statistical language models for topographic data recognition

The success of Statistical Language Models (SLMs) at improving the performance of Natural Language Processing (NLP) applications suggests their possible applicability to the area of automated map reading. This idea stems from the fact that there are similarities between natural language and cartographic language. We describe a method of using SLM to characterise the context of different classes of objects. We use these models to measure the frequency of each feature context. This can be used to help identify unclassified map features in combination with other methods (for example, based on an object’s shape).

Applications of cartographic structure matching

Automatic categorization of large-scale topographic vector data into roads, buildings and similar classes typically examines each object description in isolation. We describe a cartographic structure matching (CSM) algorithm that automatically classifies objects in topographic maps by examining the context of the object. Matching clusters of objects against known templates serves to categorize ambiguous polygons by including context in the categorization process. We describe a number of applications that emerged from the core structure-matching algorithm, addressing problems of error detection, rejoining partitioned objects, composite object identification and data quality estimation.

Data fusion for topographic object classification

This paper presents research conducted into the automatic recognition of features and objects on topographic maps (for example, buildings, roads, land parcels etc.) using a selection of shape description methods developed mostly in the field of computer vision. In particular the work here focuses on the proposal and evaluation of fusion techniques (at the decision level of representation) for the classification of topographic data. A set of Ordnance Survey large-scale digital data (1:1250 and 1:2500) was used to evaluate the classification performance of the shape recognition methods used. Each technique proved partially successful in distinguishing classes of objects, however, no one technique provided a general solution to the problem. Further outlined experiments combine these techniques, using a data fusion methodology, on the real-world problem of checking and assigning feature codes in large-scale Ordnance Survey digital data.

Graphical object recognition using statistical language models

This paper describes a proposed system for the recognition and labeling of graphical objects within architectural and engineering documents that integrates statistical language models (SLMs) with traditional classifiers. SLMs are techniques used with success in natural language processing (NLP) for use in such tasks as speech recognition and information retrieval. This research proposes the adaptation of SLMs for use with graphical notation i.e. statistical graphical language model (SGLMs). Reasoning of the similarities between natural language and technical graphics is presented and the proposed use of SGLM for graphical object recognition is described.

Statistical Language Models for Graphical Object Recognition

This paper explores automatic recognition and semantic capture in vector graphics for graphical information systems. The low-level graphical content of graphical documents, such as a map or architectural drawing, are often captured manually and the encoding of the semantic content seen as an extension of this. The large quantity of new and archived graphical data available on paper makes automatic structuring of such graphical data desirable. A successful method for recognising text data uses statistical language models. This work will investigate and evaluate similar and adapted statistical models (Statistical Graphical Langauge Models, SGLM) to graphical languages based on the associations between different classes of object in a drawing to automate the structuring and recognition of graphical data.