Bilkent University
Department of Computer Engineering


A Unified Low Level Analysis Framework for The Video Content Management Systems


Umut Naci

Delft University of Technology
The Netherlands

Low level analysis of video is a fundamental step in semantic information extraction and modeling, i.e. high level analysis of video. It consists of detecting and identifying the shot transitions (abrupt and gradual), camera motions and the scene organization (e.g. detecting sub-scenes). Although many methods specifically tailored to handle a subset of these situations have been proposed, no single method is able to provide the complete low level information in a fast and effective manner. 3D block based analysis of video data is a promising tool for extracting features that reflect the local spatiotemporal behavior of the data flow and create probabilistic models out of it to detect and identify various phenomena along the sequences. An implementation of the idea for the shot transition detection problem reflects its potential as a strong base for the complete low level analysis. Besides detecting shot transitions, the method also provides the possibility to detect the starting and ending time stamps for each transition and is highly computationally efficient.
Umut Naci is a PhD researcher at Delft University of Technology, The Netherlands. He got his BS degree in 2001 and MS degree in 2004 both from Bogazici University, Istanbul. He published papers on digital watermarking and automatic speaker recognition. His current research interests are semantic analysis of video, multi-modal analysis and content management and retrival systems


DATE: February 28, 2005, Monday @ 13:30