Each Shape description technique produce a set of real descriptor values that are used for the shape description of topographical objects
Last updated 13th February 2003
Shape, Context, and Statistical Language Modelling for recognition and validation of graphical objects
We are a sub-group of the Intelligent and Graphical Systems (IGS) research group in the Department of Computer Science, NUI Maynooth.
Adam Winstanley (academic)
Laura Keyes (research assistant)
Leo Mulhare (research student)
Bashir Salaik (research student)
Aliex Zhou (research student)
Andrea McQuillan (MCS student)
Associates
Diarmuid O'Donoghue (academic)
John McDonald (academic)
(Re-)Structuring topographic data for GIS
Validation and Quality assessment of Geographic Data
Recognition of complex objects with graphical data
Structuring data for multi-media systems
Projects
Topographic Shape Recognition
Structure MatchingTopographic Shape RecognitionStructuring Archaeological Features on Topographic Digital maps
Recognition, Labelling and Retrieval of Features on Technical Drawings
Recognition of objects is largely based on the matching of descriptions of shapes. Numerous shape description techniques have been developed in computer vision and image processing such as boundary chain coding, analysis of scalar features (dimension, area, number of corners etc), Fourier descriptors and moment invariants.. A comparison is made of the effectiveness of each method for recognising features on large-scale topographic maps and plans.
The above techniques are evaluated as general classifiers applied to broad classes of topographic shape (buildings, fields, road etc.) using the sample data provided by Ordnance Survey (OS) Great Britain. One of the aims of this project is to exhaustively test each of the methods and to perform a statistical analysis on the range of descriptor values obtained both within and between each OS feature type. For example, to evaluate how the techniques distinguish buildings from land parcels but also how they distinguish each type of land parcel i.e. surface land parcel or defined natural land-cover. Another aim is to evaluate the classification performance of each method on all polygons through comparison with the original data and to compare the performance between the three methods.
When tested for the more generalised topographic shapes, Fourier descriptors do not appear to be as conclusive and successful as hoped. However, both the moment invariants and scalar techniques proved to be significantly more successful in their task. Results show that no one shape technique alone is powerful enough for the task i.e. in different situations one technique will perform better than the others and produce significant results (e.g. buildings from linear features in built up areas using the moment invariants technique).
Keyes, L and Winstanley, AC: Using Moment Invariants for Classifying Shapes on Large Scale Maps, Computers, Environment and Urban Systems 25(1), 119-130, January 2001.
L. Keyes and A.C. Winstanley: Data Fusion for Topographic Object Classification, IEEE/ISPRS Joint Workshop on Remote Sensing and Data Fusion over Urban Areas, 8-9 November 2001.
Keyes, L and A.C. Winstanley: Topographic Object Classification Through Shape, GIS Research UK, 383-387, Glamorgan, April 2001 (extended abstract).
L. Keyes and A.C. Winstanley: Using Moment Invariants for Classifying Shapes on Large Scale Maps, IMVIP 2000 Proceedings of the Irish Machine Vision and Image Processing Conference, 149-156, Queen's University Belfast, September 2000.
A.C. Winstanley, and L. Keyes: Applying Computer Vision Techniques to Topographic Objects, XIXth International Archives of Photogrammetry and Remote Sensing, 33 (B3), 480-487, July 2000.
Keyes, L and Winstanley, AC: Moment Invariants as a Classification Tool for Cartographic Shapes on Large Scale Maps, 3rd AGILE Conference on Geographic Information Science, Helsinki, May 2000.
Keyes, L and Winstanley, AC: Using Moment Invariants for Classifying Shapes on Large Scale Maps, GIS Research UK, York, April 2000 (extended abstract).
L. Keyes and A.C. Winstanley: Fourier Descriptors as a General Classification Tool for Topographic Shapes, IMVIP 1999 Proceedings of the Irish Machine Vision and Image Processing Conference, Dublin City University, 1999.
L. Keyes and A.C. Winstanley: Topographic object recognition through shape, NUIM Signals and Systems Research Group, Technical Report NUIM/SS/--/2001/06, 2001.
Parallelisation of algorithms for analysing graphical data, (with Clemson University) Enterprise Ireland International Collaboration Programme IC/2000/082, IR£1300 (2000-1)
Topographic Object Recognition (with D. O’Donoghue and Ordnance Survey), Enterprise Ireland/British Council Research Visits Scheme BC/2002/015, €2500 (2002)
Structuring topographic data through object shape, Ordnance
Survey (GB) Research Grant, UK£1000 (2000-1)
Classification Error Detection
Error detection is central to the quality assurance needs of national
ordnance survey offices. Specific classification errors are identified
by explicitly defining an illegal context - for example, a section of road
that does not border another road segment. Detecting specific errors is
perhaps of greatest use when there is a known problem with an exexisting
categorisation processes.
Quality Estimation/ frequency Distribution
Previous results indicate there is an exponential distribution in the
frequency with which different contexts occur within a map. However, individual
map segments may vary, perhaps using more urban related contexts. When
updating a map (segment) we may compare the distribution of context before
and after update - any significant discrepancy may indicate an error in
the updating process.
Rejoining Segmented Objects
Topological data is a two-dimensional (2D) representation of three-dimensional
(3D) information. Occlusions frequently segment objects, like bridges occluding
the underlying river. CSM can identify such contexts, and introduce an
occluded object segment.
Composite-Object Identification
Topological data is stored as individual land parcels. Introducing
hierarchical structure based thematically related collections of objects.
Such collections are generally adjacent, and thus CSM is ideally suited
to identifying such structures. For example, a road plus adjacent
Mulhare, O'Donoghue, Winstanley Context-based classification of objects in cartographic data, GISRUK Geographical Information Science Research Conference pp195-198, Sheffield, UK, April 2002.
Mulhare, D. O'Donoghue, A. C. Winstanley, Analogical Structure Matching on Cartographic Data, 12th Artificial Intelligence and Cognitive Science AICS-2001, NUI Maynooth, Ireland, pp 43-53, Sept. 5-7, 2001, ISBN 0-901519-48-0.
O'Donoghue, D., Adam Winstanley., Finding Analogous Structures in Cartographic Data, 4th AGILE Conference on G.I.S. in Europe, Czech Republic, April, 2001.
Adam Winstanley, Diarmuid O'Donoghue, and Laura Keyes, Topographical Object Recognition through Structural Mapping, 1st International Conference on Geographic Information Science - GIScience 2000 -, Savannah, Georgia, USA, October 28-31, 2000.
Bohan, O'Donoghue A Model for Geometric Analogies using Attribute Matching, AICS-2000 11th Artificial Intelligence and Cognitive Science Conference, Aug. 23-25, NUI Galway, Ireland, 2000.
Topographic
Object Recognition (with D. O’Donoghue and
Ordnance Survey), Enterprise Ireland/British Council Research Visits Scheme
BC/2002/015, €2500 (2002)
Context sensitive categorisation of topographic data, Ordnance
Survey (GB) Research Grant, UK£500 (2000-1)
The success of statistical language models 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:
There are, of course, also many differences between the two
forms of data, notably the one-dimensional sequence of words forming a national
language text compared with the two-dimensional graphical map. Natural language
also has a large vocabulary whereas the number of classes of topographic object
is usually small.
Publications
The work involved in converting large data-sets into an object model is considerable. Therefore, it is very labour intensive to structure the data manually. Some automation of this process is possible. As an extension to previous work on object recognition in topographic data this project aims to implement an automatic procedure for the identification, extraction and depiction of archaeological features on large-scale maps. The system searches the data-sets for likely archaeological features and confirms their status as such. It also ascertains their extent by distinguishing between features that do and do not belong to the site. Complications will arise because top and bottom of slope may also represent modern man-made features such as embankments or road cuttings. Therefore it is important to distinguish the actual archaeological features from other anthropogenic forms. Our previous work carried out on object recognition is applied to this problem. Then a bounding polygon is created using either existing geometry and/or new lines and a composite model is formed to include the relevant features and geometry. A hierarchical structure to represent the site is then built.
Outline of process involved in developing the tool in ArcView
For many applications, a generalised representation of the archaeological site is required. This is generated using geometric techniques such as the medial axis transform and the crust method of curve reconstruction. This automatic generalisation of the archaeological site produces a simplified version of the site plan suitable for small-scale depiction, morphological analysis and reconstruction of the original configuration of the site.
Keyes, L and A.C. Winstanley: Automatic Identification of Archaeological Features on Digital Topographic Maps, NUI Maynooth Postgraduate Research Record: 2002
A.C. Winstanley and L. Keyes: Identification, Extraction and Depiction of Heritage Features in OS MasterMap Data, Technical Report, Department of Computer Science, NUIM, 2002-2003.
Funding
Recognition and Structuring of heritage features in topographic data, Ordnance Survey (GB) Research Grant, UK£3000 (2001-2)
The information is accessed through a standard web browser interface including navigation through hot-links and key-word search facilities. The CAD drawings showing the location of utilities and services also act as browser navigational maps. In operation, the system’s main use concerns day-to-day operation and maintenance tasks, for example:
Funding
Automatic
recognition and labelling of features on technical drawings (with
Entropic Ltd, Maynooth), Enterprise Ireland Innovations Partnership Feasibility
Study IP/2002/064, €9000 (2002)
What quality measurements need to be considered and how are they
assessed
The causes of error in spatial data may come from varied sources, such
as, understanding and modelling of reality, the source data and data encoding,
conversion, analysis and output. To control the affects of error on spatial
data the following data quality aspects need to be measured and assessed.
The actual quality checking required may vary from application to application
and the context of use. An example of the relationship between the quality
indicators and the model used to measure them can be seen below (this is
not a final model and will be modified to include more detailed quality
metrics).
Relationship between the quality components and the quality metrics.
GI Quality measurement, assessment and representation
As a group (IGS) we propose to produce a prototype tool-set, which uses a set of defined feature quality indicators to detect and quantify spatial data quality errors on topographic maps and databases. Firstly, methods of measuring quality related errors, and means determining appropriate quality indicators dealing with these errors need to be developed. Shape and context recognition has already proven useful in the detection of misclassified features. It is envisaged that work, previously done on shape and context recognition both as individual and combined procedures can be directly applied in some form to many of the issues involved in data quality. These include completeness of cover, attribute accuracy, logical consistency and misclassification rates. We plan, through other statistical and data descriptor methods borrowed from such fields as computer vision and engineering to introduce new techniques to act as quality metrics to resolve all issues of quality. The following steps outline the most significant aims of our research.
Tasks