User Interface Design & Evaluation for a Content-Based Image Retrieval System.
Colin C. Venters
DepartmentofInformation&LibraryManagement,
UniversityofNorthumbria atNewcastle,
LipmanBuilding, Sandyford Road,
NewcastleUponTyne,
NE18ST, England.
RESEARCH BACKGROUND
The developing technology of content-based image retrieval (CBIR) is creating new and exciting opportunities to enhance the access to and retrieval of digitally stored images. Paradoxically, while the underlying technology of CBIR systems is being advanced both system developers and researchers have generally overlooked the importance of the human-computer interaction (HCI) and the crucial role of the user interface. For example, the ability to formulate and communicate a query is an essential feature in all information retrieval systems. However, a number of CBIR systems provide only primitive facilities for image query formulation, or restrict end- users to a single visual-browsing query option to interrogate the database of stored images. The major problem of designing a user interface for CBIR systems lies in the need to provide users with appropriate interface features and tools to facilitate HCI, yet sound design techniques for facilitating retrieval are rare. It is the aim of this doctoral research programme to address this problem. This research project aims to design and evaluate a user interface for a CBIR system using the ARTISAN system as an example.
RESEARCH TO DATE
Stage I: End-User Group Identification
Potential groups of end-users were selected on the basis of suitability to utilise the ARTISAN system. UK trade mark and patent companies, patent information network (PIN) libraries, and patent services were identified and selected using a combination of both commercial references sources and professional directories. This enabled the development of a substantial database of both private practice firms and personnel in the UK. In total, 144 companies and 316 potential volunteers were identified. A criterion for selecting volunteers had been developed prior to a postal mail shot in order to provide a cross-section of end-users'. This criterion was designed to provide an insight into specific areas of potential volunteers' experience and focused on a number of important areas: the number of years in the profession, level of computer literacy, user interface familiarity, image retrieval system use, and perceived level of image retrieval skill. Users were requested to complete background questionnaire in order to assess their suitability.
Stage I: User Requirements Gathering
A tape recorded, semi-structured interview format was selected as the method by which the user requirements would be gathered. An interview framework was then developed based on an object-oriented diagram. This provided structure and flexibility allowing both end-users and the interviewer to explore salient issues. Twenty-three interviews were then conducted between August-September 1997 with the targeted group of end-users. The aim of the interviews was to explore in-depth the user
requirements for a CBIR system in general, and the ARTISAN retrieval system in particular. A number of key areas and issues were explored within the topics of system input, system output, and user interface issues.
Stage II: User Requirements Analysis
No perceived benefits would be gained by the full transcription of the 23 interview tapes, as a result the data collected from the interviews was partially transcribed to enable analysis of the data. The data collected during the user requirements process highlighted a number of functional and data requirements common to the overall system, and a number of system features necessary for visual query formulation. From the data gathered 5 methods were proposed for visual query formulation: a pre- defined graphic file, a scanning tool, a free-hand drawing tool, a visual browsing tool, and a visual building tool. Contrary to current system features, these findings suggest that end-users have a range of query needs, which can be supported by the identified query formulation tools. Although the end-user groups have a shared interest in the effective retrieval of the dataset, the findings highlighted a distinction between the user requirements of trademark searchers and that of PIN information officers. This can be attributed to the type of client and nature of the service they provide. The analysis of the data collected from the interviews has resulted in the production of a number of abstractions of the system: dataflow diagrams, entity relationship diagram, object-oriented model, and a viewpoint-oriented model. These findings and abstractions have been evaluated with a small cross-section of the initial end-user population who validated the findings. As a result of the requirements validation process these identified features will form the basis for the conceptual and holistic design of the user interface, and the development of a horizontal prototype in the next stage of the project.
CONCLUSION
To assure effective and efficient access and retrieval, the design of the user interface for CBIR systems needs to be dramatically improved. If CBIR systems are to become viable applications, the design of a suitable user interface for query formulation and the manipulation of search results is a fundamental aspect of developing the systems in conjunction with the underlying technology. The outcome of this research will contribute a framework for the design and development of more user-centered interfaces for content-based image retrieval systems.
