Multi-agent Image Retrieval System Based On Jade

In the 21st century, with the development of digital image acquisition devices, network technology and communication technology, the number of digital images grows fast. But how to find the needing images quickly from the database has become a serious problem. The traditional text-based image retrieval technology could not satisfy the requirements of the rapid growth of image number due to its inherent disadvantages. Content-based image retrieval technology could overcome the shortcomings of text-based Image retrieval; however, the gap between low-level visual feature and the high-level semantic limits its development in practical applications.In this paper, multi-Agent technology, the most important branch of artificial intelligence, is used in content-based image retrieval. The agent cooperation, control, communication and interaction are used to improve the intelligence of image retrieval. This paper designs and develops a multi-agent image retrieval system based on Jade which is developed by Telecom Italia department. The system is composed of several independent agents and each search agent encapsulates one or several image retrieval algorithms and has the ability of independent retrieval. Retrieval agents interact with users through integration agent. Users work as an intelligent agent that has the high-level semantic knowledge and guide the actions of retrieval agents. The system adopts hybrid structure. Agents register to the AMS Agent which manages the multi-agent platform and lifecycles of every agent when it starts. The DF agent issues the information of services description and searches services. The communication paradigm between agents is based on asynchronous message passing model. The message receiver is conformed by the global domain name. The FIPA-ACL which is used in communication separates the communication actions and content language. Communication actions direct the action of agents and the content language describes the content of information. Agents cooperate with each other to accomplish the task. The relevance feedback method is used in the system to improve the subjectivity of retrieval. Moreover, Search agents improve retrieval parameters that satisfy the needing of user according to the images which uses selected. The system allows for dynamic addition of search agents incorporating new image retrieval technology which improve the extensibility of system.The result shows that the multi-agent image retrieval system which we design performs well in the search of natural images which are classified in vehicle, landscape, animals, flowers and fruits. The study of this paper will lay the foundation for promoting the combination of image retrieval technology and the artificial intelligence technology, solving the semantic gap of content-based image retrieval and improving the intelligence of image retrieval.