Clustering de visages : vers la construction automatique d'un album photo à partir d'une s equence vidé o

Abstract : This paper presents a clustering method of detections of the same object seen on a video. We apply it to the context of the automatic construction of photo album. We use a global analysis, based on a probabilistic framework of data association problems. The solution is given by Maximum A Posteriori estimation. Our main contribution concerns the use of a local front-back tracking, applied to each detection ; to increase appearance information of detections with a spatial information, through local tracks construction. We introduce a new likelihood measure based on the spatio-temporal dissimilarity between tracks. The algorithm is then able to deal with situations in which the face detections are scattered. We propose to use criteria derived from purity and inverse purity of a clustering to assess performances of the proposed method. This method is compared to hierarchical clustering on two real test sequences.
Document type :
Conference papers
Complete list of metadatas

Cited literature [14 references]  Display  Hide  Download

https://hal.inria.fr/inria-00595741
Contributor : Peter Sturm <>
Submitted on : Wednesday, May 25, 2011 - 2:24:34 PM
Last modification on : Tuesday, July 2, 2019 - 4:08:03 PM
Long-term archiving on : Friday, November 9, 2012 - 12:10:32 PM

File

paper28.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00595741, version 1

Citation

Siméon Schwab, Thierry Chateau, Christophe Blanc, Laurent Trassoudaine. Clustering de visages : vers la construction automatique d'un album photo à partir d'une s equence vidé o. ORASIS - Congrès des jeunes chercheurs en vision par ordinateur, INRIA Grenoble Rhône-Alpes, Jun 2011, Praz-sur-Arly, France. ⟨inria-00595741⟩

Share

Metrics

Record views

165

Files downloads

164