Content-Based Indexing and Retrieval Supported by Mobile Agent Technology [chapter]

Harald Kosch, Mario Döller, László Böszörményi
2001 Lecture Notes in Computer Science  
In this paper we present the MultiMedia Database Mobile agent technology (M 3 ) which supports personalized content-retrieval and indexing in a distributed Oracle 8i DB. We implemented an agency on top of the Oracle 8i JServer and realized mobility with the embedded Visbroker Corba ORB. A performance comparison of our mobile agent technology with a client-server solution for a nearest-neighbor search in an image database shows the efficiency of the proposed solution. similar features to the
more » ... ications mentioned beforehand : they are characterized by asynchronous transactions, high latency, complex information processing and distributed task processing features. Related Work One key functionality in a multimedia database (MMDMBS) is how to index and then to retrieve continuous and non-continuous multimedia information efficiently. One broadly used method, the Content-Based Retrieval (CBR) of multimedia objects relies on extraction properties of a multimedia object [7, 8] . CBR in distributed MMDBMS involve the retrieval of multimedia data from various, possibly heterogenous database sites, and to compute the result in mutual agreement. A typical approach to CBR is the similarity search between extracted multimedia features [7] . Here, the query is actually posed against the low-level feature vectors extracted from the multimedia object. A broadly used Image CBR system is QBIC (Query By Image Context, see wwwqbic.almaden.ibm.com), another popular system for videos is Virage (www.virage.com). Features described for CBR include measures expressing the color/texture/shape distribution of an image plus motion for a video. A CBR query is translated into a point query in a multi-dimensional feature space. The similarity between a query-and a database-object is estimated using a distance function. CBR in a distributed MMDBMS is broadly supported by a client/server architecture. It includes (1) user interfaces to submit requests, their transfer to the DB server; (2) the retrieval operations at the DB server; and (3) the results return. A broadly used protocol for the client (written in Java) and server is JDBC (Java Database Connection; see java.sun.com/products/jdbc/index.html). If several sites are involved, multiple client-server connections are spawned and the different results are compared and merged at the client-side. An example is MetaSEEk (www.ctr.columbia.edu/MetaSEEk/). It is a meta-search engine for images based on the content of CBIRs located at IBM, Virage and Columbia University servers. Our previously developed SMOOTH system [9] provides besides CBR also means for high-level annotation and querying, works in a client-server environment as well. A recent system enhancement (introduction of 'Domain Transparency' meaning that the client interface automatically adapts to extensions made by a new application domain to the base annotation classes) revealed serious performance bottlenecks of the JDBC connection (thin driver) to an Oracle 8i DB. The repeated JDBC calls to build the client interface dynamically combined with a high volume of requested data worsened the response time considerably. A first-aid solution was the integration of a JDBC client cache for query results. However, the use of a mobile agent solution might improve the situation more efficiently. We address such a solution in the near future. Many mobile agent systems have been established. Some of these are Aglets [10], Mole [11] and Grasshoper [12] . They are well-suited to a wide range of Internet applications, to mention only WWW mining [13], or telecommunication [14] .
doi:10.1007/3-540-44819-5_13 fatcat:xr3nyj277bdatfsdqhiwcy57ga