Using multiple agents for content-based image retrieval

In this paper we describe a hybrid system for dealing with content-based image retrieval. Lilies is used in conjunction with a neural network to match and retrieve images. We explain how we deene and use our image indexes which allows us to treat the problem in a distributed fashion. We combine the indexes with a multiple agent approach to help reduce our search space and to retrieve an image as a parallel action. The resulting system is one which is exible but controlled. A neural network is used to generate a feature vector from the images which are used by the agents. The agents are based on the theories behind Lilies. Lilies (Localisation and InterLeaving stragIES 4, 5, 3]) was developed to deal with multiple agents for planning environment. The application required adaptive planning in a reactive and generative environment. To model the application, it was necessary for multiple agents to be developed which added the usual communication 14, 19] issues. To provide a uniied and adaptative environment we used reeection 20, 12, 3, 18]. Each agent in Lilies is like a small blackboard 8], similar to CASSANDRA 6]. However, Lilies is agent based and provides a heuristic adequate structure for the a planning environment which requires adaptive properties. We have applied this multiple agent strategy to dealing with content-based image retrieval problems for image recognition. In this paper we will explain the structure of lilies and how to apply this multiple agent theory to content-based image retrieval. Although, using the blackboard for image recognition has been considered inadequate, through the extensions made by Lilies and the organisation of the application we are able to overcome the multiple dimension properties of content-based image retrieval.