Efficient multimedia retrieval using custom indexing methods

• Main Author: Steenburgh, Malcolm W.
• Format: Others
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/9059

One of the largest problems associated with content-based indexing of multi-media documents is the inefficiencies associated with the process. These inefficiencies are witnessed in a number of areas throughout the process, manifesting themselves in the form of high network bandwidth and processor requirements. The following piece of work examines two novel concepts which attempt to minimize these inefficiencies through the reduction in both bandwidth and processor requirements. Included in this work is a discussion related to the reduction of network bandwidth requirements. The discussion focuses around the network bandwidth currently associated with the development of indexing schemes. In order for developers to implement and test indexing schemes, they currently have to download and index all of the media to which the new scheme will be applied. In addition to this being time consuming and work intensive, the entire process results in large amounts of network traffic. The proposed solution to this problem is a software interface allowing developers to submit unique indexing schemes to remote servers for processing. Upon receipt, the remote server applies the scheme to an arbitrary subset of its indexed media, restricted only by a set of time constraints set by either the developer or server administrator. Within the time constraints, the server will apply the given scheme and return the best results to the user. The result is efficiencies that are orders of magnitude greater than possible through current methodology. The second concept discussed in the work focusses on the efficient application of known indexing schemes in query by example format queries. The algorithm makes use of the triangle inequality to reduce the number of media objects that actually have to be compared during query processing. In addition to a disussion of the aforementioned concepts, this work includes an examination of an actual implementation of the concepts using the Java programming language. The application shows the viability of both concepts, showing that they indeed reduce the inefficiency currently associated with multi-media indexing systems. === Science, Faculty of === Computer Science, Department of === Graduate