Automatic Topics Segmentation for News Video
by Clustering of Histogram of Orientation
Gradients Faces
Mounira Hmayda, Ridha Ejbali,
and
Mourad Zaied
RTIM: Research
Team in Intelligent Machines,University of Gabes, National Engineering School of Gabes
(ENIG), Tunisia
Abstract: TV
stream is a major source of multimedia data. The proposed method aims to enable
a good exploitation of this source of video by multimedia services social community,
and video-sharing platforms. In this work, we propose an approach to the
automatic topics segmentation of news video. The originality of the approach is
the use of Clustering of Histogram of Orientation Gradients (HOG) faces as
prior knowledge. This knowledge is modeled as images which governs the
structuring of TV stream content. This structuring is carried out on two
levels. The first consists in the identification of anchorperson by
Single-Linkage Clustering of HOG faces. The second level aims to identify the
topics of news program due to the large audience because of the pertinent
information they contain. Experiments comparing the proposed technique to
similar works were carried out on the TREC Video Retrieval Evaluation (TRECVID)
2003 database. The results show significant improvements to TV news structuring
exceeding 96 %.
Keywords: Anchorperson,
clustering, face detection, features extraction, news program.
Received December 28, 2018; accepted April 10, 2020