CHI 97 Electronic Publications: Demonstrations
Depictive Interaction with Visual Information Using Sketches -DIVIUS
Andree Woodcock
Design Research Centre
Derby University, Derby, UK
email A.Woodcock@derby.ac.uk
Stephen A. R. Scrivener
Design Research Centre
Derby University, Derby, UK
email S.A.R.Scrivener@derby.ac.uk
Mark Lansdale
Department of Human Sciences
Loughborough University, Loughborough, UK
email M.W.Lansdale@lboro.ac.uk
ABSTRACT
Querying of visual databases has relied predominantly on text based systems. Words do not provide an appropriate or adequate means of describing visual artifacts. A system (DIVIUS) has been developed which allows users to describe and query objects in a pictorial database, using a visual language derived from the database objects. Users can also indicate their level of uncertainty regarding certain attributes of the query.
KEYWORDS
Visual interface, pictorial database, uncertainty, database evaluation, user models.
© 1997 Copyright on this material is held by the authors.
INTRODUCTION
A pictorial or visual database is a series of images which have been amassed for a particular reason. They can include photographs, drawings, paintings, slides, prints, or X-Rays. Visual databases may be interrogated for many reasons such as instruction, communication. exemplification, sourcing, problem solving. In many cases the only way in which material can be accessed is through a text based classification system, developed primarily to aid cataloguing the material, not selective retrieval. For example, a doctor might remember having seen a similar cell pattern before, but would be unable to locate this particular slide if items were stored solely by patient number.
Text based encoding (or classification) of visual databases restricts the likelihood of mediating a successful search (if the search relies on visual memory), and limits the usefulness of the database.
DIVIUS (Depictive Interaction with Visual Information Using Sketches) was developed as a visual interface to a pictorial database. Using the same interface, an expert/curator can encode/classify items in the database, and a user to retrieve items from it. As both expert and retriever use the same 'restricted' visual language some of the problems of database searching are assuaged.
DIVIUS was developed to explore three issues: firstly could visio-spatial information be used to search for items in a database: secondly, is a schematic (identikit type representation) an appropriate level of representation; and thirdly can any added value be brought to the search by using uncertainty.
THE SUBJECT DOMAIN
Ceramics was chosen as the domain to test out the interface. A computer database of 120 cups of the period 1790-1810 was developed. From text based descriptions of the artifacts, and the artifacts themselves, a visual language consisting of 9 features and their related attributes was developed (see Figure 1). This visual language was used for classification and retrieval.
DESCRIPTION OF SYSTEM
DIVIUS was developed in Delphi on a PC Pentium [1]. The interface is shown in Figure 1. It consists of a 'sketch area' where the schematic representation is built up from items selected from the high level menus and their associated palettes. The size and position of all items can be changed in the drawing area and are stored as part of the encoding of that object. All the visual information to be contained is input from this interface.
The same interface is used to retrieve an item from the database. As the user might be specifying a query from memory, levels of uncertainty with regard to the position [2] of borders, fluting and handle can be entered as well. Once the query has been completed the user can select any or all of the attributes to be included in the search.
The results of the query are presented to the user first in the form of a match matrix [3]; a 16 level grey scale matrix showing the nearness of the match on each attribute to items contained in the database. An overall match rating is also given. In the present configuration, only those items which achieve a perfect match in terms of the high level features are displayed. Clicking on the cells will pull out a thumbnail image from the database.
EVALUATION
The system was evaluated by 8 postgraduate students at Loughborough University. It was found to be both easy to use and enjoyable. Subjects were able to use the schematic representation to identify target objects, and could construct a query to retrieve the target item from the database. The accuracy with which visio spatial information was used, and recalled by subjects was high. All subjects thought the representation of uncertainty was desirable, but did not like the way it had been implemented
CONCLUSIONS
It is increasingly easy to compile pictorial databases. However, little attention has been given towards how users might go about retrieving items from them. Most current methods rely on text based entry or presentation of a large number of items to the user (eg browsers). Generating text based descriptions of visual material continues to be costly (in terms of labour, training and thesaurus development) and not always appropriate.
DIVIUS uses the same visual language for encoding and retrieval. On average, it took 4 minutes to encode an item in terms of 40 visio -spatial features. Although the language itself is domain specific it is hoped the approach itself will be extensible.
DEMONSTRATION
The demonstration will focus on showing the way in which items can be encoded and retrieved from the database, using the sketch area, uncertainty measures, attribute lists and match matrix. It is hoped that this will provide opportunities for the discussion of wider issues of pictorial database design, and user centred approaches to retrieval interfaces.
ACKNOWLEDGEMENTS
The research was conducted in the Departments of Human Science and Computer Studies at Loughborough University, supported by the MRC/ESRC/SERC Joint Council Initiative, grant number G9200538
REFERENCES
- Bartlett, R.A., The development of the DIVIUS Database, Project Report, Loughborough University, 1996
- Lansdale, M. W., Modelling errors in the recall of spatial position, MS submitted for publication, 1996
- Lansdale, M.W. and Edmonds, E.A., Using memory for events in the design of personal filing systems, IJMMS, 36, 97-126, 1992.
- Woodcock, A., Scrivener, S.A.R. and Lansdale, M.W., User requirements of visual interfaces to pictorial databases (in press).
Figure 1: The DIVIUS interface showing a schematic representation of the object
CHI 97 Electronic Publications: Demonstrations
