Audio-Video Person Clustering in Video DatabasesReport as inadecuate




Audio-Video Person Clustering in Video Databases - Download this document for free, or read online. Document in PDF available to download.

Martigny, Switzerland: IDIAP, 2003

A methodology towards person clustering in meeting databases is presented in this report. Such goal is generic to a number of problem in computer vision and more specifically in content-based video indexing and retrieval. First, the audio-stream was considered alone, leading to the speaker clustering problem. An already existing algorithm has been used to this end. Then, the video stream was analysed, leading to a face clustering algorithm build from probability density distribution of face similarity distances. Finally, both modalities are considered to combine voice clustering and face clustering and achieve person clustering. This method has been tested on the IDIAP meeting database, and many results are given to prove the efficiency of the method and to show that it can be applied to other databases.

Keywords: vision ; odobez Reference EPFL-REPORT-82953





Author: Kottelat, F.; Odobez, Jean-Marc

Source: https://infoscience.epfl.ch/record/82953?ln=en







Related documents