Abstract
Recent computer advances have made it possible to store graphic and pictorial information in large quantities. The problem is how to provide easy and effective access to this material for the casual user. This paper presents an enhancement to our feature-matching approach to graphic retrieval. In an earlier paper, we proposed a feature-matching approach in which users describe what they want to retrieve in response to a set of queries. The user's description is then matched by our system with descriptions of pictures in the database. Database pictures can then be presented in order of similarity to the user's description. The change in our approach is to scale the presence or absence of each feature on a continuum from clearly present in a picture to clearly absent. Reanalysis of our empirical data for a database of 1016 pictures indicate a significant improvement in retrieval performance (mean number of pictures which must be examined to find the target = 3.0). More importantly, the worst case was reduced from 208 to just 20.