Local Features for Mining Multimedia Data and Connecting the Real and Digital World

Till Quack

Information Technology and Electrical Engineering
ETH Zurich

Date: Friday, October 26, 2007
Place: Buchanan 1940
Time: 2:00 pm — 3:00 pm


Abstract:

The talk will consist of two parts. The connecting element between the talks is the use of local visual features for object recognition.

In the first part we present an application of the traditional itemset mining techniques in the non-traditional domain of visual data. We mine configurations of local features which are characteristic for a class of objects, such as cars, bikes or giraffes. Frequent feature configurations are mined from training data for each object class. Based on the mined configurations we develop a method to select features which have high probability of lying on previously unseen instances of the object class.

The second part of the talk will deal with object recognition from mobile devices such as cellphones. This allows for interactive multimedia applications such as interactive newspapers, interactive meeting rooms, or cityguides. This application has reached commercial feasability and has led to the foundation of a spin-off.

TILL QUACK received his Masters degree in Information Technology and Electrical Engineering from ETH Zurich, Switzerland in 2004. At the time he conducted his Masters project with Dr. Manjunath at UCSB through an exchange program between ETH Zurich and UCSB. Currently he is a PhD student in Luc Van Gool's group at ETH Zurich and co-founder of the spin-off kooaba.

Host: Prof. B.S. Manjunath, Electrical and Computer Engineering