A comparison of clustering algorithms for face clustering

Bijl, Erik (2018) A comparison of clustering algorithms for face clustering. Bachelor's Thesis, Computing Science.

Preview	Text Report_research_internship.pdf Download (1MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (94kB)

Abstract

Video surveillance methods become increasingly widespread and popular in many organizations, including law enforcement, traffic control and residential applications. In particular, the police performs inves- tigations based on searching specific people in videos and in pictures. Because the number of such videos is increasing, manual examination of all frames becomes impossible. Some degree of automation is strongly needed. Face clustering is a method to group faces of people into clusters containing images of one single person. In the current study several clustering algorithms are described and applied on different datasets. The five clustering algorithms are: k-means, threshold clustering, mean shift, DBSCAN and Approximate Rank-Order. In the first experiments these clustering techniques are applied on subsets and the whole Labeled Faces in the Wild (LFW) dataset. Also a dataset containing faces of people appearing in videos of ISIS is tested to evaluate the performance of these clustering algorithms. The main finding is that threshold clustering shows the best performance in terms of the f-measure and amount of false positives. Also DBSCAN has shown good performance during our experiments and is considered as a good algorithm for face clustering. In addition it is discouraged to use k-means and for large datasets Approximate Rank-Order when clustering faces.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Sobiecki, A. and Wilkinson, M.H.F.
Degree programme:	Computing Science
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	26 Jul 2018
Last Modified:	27 Jul 2018 12:35
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/18064

Actions (login required)

View Item