1. Image Retrieval

Content-based image retrieval is used for retrieving historical patterns similar to those that users are interested in. The motivation for retrieving similar patterns archived in the database is that past instances of typhoons give some information on the analysis of the current instance. If we could perform an "instance-based prediction" based on past similar patterns, it would be interesting; howver, this kind of prediction has received very pessimistic outlook in the meteorology community. The reason for this is the chaotic nature of the atmosphere. However, the representation of typhoon shape, and the definition of similarity between typhoon shape has still much to be investigated.

2. System Architecture

System Architecture We are building IMET (Image Mining Environment for Typhoon analysis and prediction), which is designed for the intelligent and efficient searching and browsing of the typhoon image collection. The figure illustrates the overall architecture of IMET. The system consists of three main components: Web browser clients as the user interface, the Web servers which may act as hierarchical meta servers, and backend database servers where typhoon data are actually archived. Moreover, those components may be distributed over the network to allow distributed database systems, which is often the case with satellite data archives. At the moment, we have two types of backend database servers, namely relational database management systems (RDBMS) and our hand-crafted image search engine called FSE (Feature Space Explorer).

For this architecture, we need to prepare two types of languages --- namely, a query language and a defition language. For our hand-crafted image database engine, we are also creating our own query language, which relies on XML for the syntax of the language, and also relies on XQuery (Query Language for XML), or its full XML-encoded XQueryX (XML Syntax for XQuery), and other XML-related standards for the semantics of the language. The advantage of having such hand-crafted languages is in rapid prototyping of new tasks required for typhoon data mining. Our intention is not in developing full-fledged languages with rigorous theoretical foundations. Instead, the design goal of our languages is to create a handy yet useful languages with maximally orthogonalized operators whose combination describe various actions needed in IMET.

3. Demonstration