Guest Speaker

MULTIMEDIA KNOWLEDGE: DISCOVERY, CLASSIFICATION AND NAVIGATION

Ana B. Benítez
Columbia University

Tues., Mar. 16th
3:30, Dreese Labs 480
Refreshments will be served at 3:18 in the lecture room.
All interested parties are invited.

ABSTRACT:
The important proliferation of multimedia such as annotated images requires advanced techniques for representing and discovering useful knowledge from multimedia for intelligent organization, navigation and retrieval.

Humans seem to build models of the real world containing both semantic and perceptual knowledge. For example, we know that cats are animals (semantic) and that they have soft fur (perceptual). In this talk, I will introduce a unified knowledge representation framework, MediaNet, the first of its kind to use multimedia for representing semantic and perceptual information about the world.

I will also present the proposed techniques for discovering perceptual and semantic knowledge from annotated images, and techniques for summarizing and evaluating multimedia knowledge. In contrast to prior work, the proposed knowledge discovery techniques integrate both the processing of images and annotations. Furthermore, my summarization and evaluation techniques are automatic, generic and applicable to any multimedia concept network.
Finally, I will describe novel ways of exploiting automatically extracted multimedia knowledge, including knowledge from external resources such as WordNet, to enhance the classification and navigation of images. Existing techniques rely solely on features directly extracted from the multimedia so they are limited by capabilities of current media analysis tools.

Host: Donna Byron