Charla: Introduction to Similarity Search in Multimedia Databases

Charlista: Tomas Skopal
23 Marzo, 2011 - 12:00
Auditorio DCC

In this lecture we give an introduction to the wide topic of similarity search in multimedia databases. Unlike table rows in a relational database, the objects of a multimedia database (i.e., images, audio, or video documents) have no explicitly defined structure of their content. Furthermore, unlike text documents or web pages that are composed of words, the multimedia content has no precise semantics. For example, a web page about car racings could be easily queried by a keyword query like "Ferrari AND winner". On the other hand, a collection of (not annotated) pictures taken on a car racing event cannot be searched so easily for "Ferrari" (or even "winner"), because the spots of red pixels possibly forming a Ferrari in the image is a knowledge that is understandable only to an experienced human. Therefore, searching for multimedia objects by its raw content is much more challenging task than searching a relational database, a text collection, or the Web. As a particular solution to this problem, the paradigm of content-based similarity search has been developed over the last two decades. We present the basic steps of the similarity search process: the feature extraction and similarity modeling, the querying, and finally the indexing for fast search.

Secretaría de Investigación