University of Cambridge Face Database

Preface

The University of Cambridge Face Database is a collection of video sequences of largely unconstrained, random head movement in different illumination conditions, acquired for the purpose of developing and evaluating novel face recognition algorithms. This page describes the database in detail.

Publications

This is a list of the main publications in which the Cambridge Face Database was used:

	T-K. Kim, O. Arandjelović and R. Cipolla. Boosted manifold principal angles for image set-based classification. Pattern Recognition, 2006. [LINK]
	O. Arandjelović and R. Cipolla. Face recognition from video using the generic shape-illumination manifold. In Proc. IEEE European Conference on Computer Vision, 2006. [LINK]
	O. Arandjelović and R. Cipolla. A new look at filtering techniques for illumination invariance in automatic face recogition. In Proc. IEEE Conference on Automatic Face and Gesture Recognition, 2006. [LINK]
	O. Arandjelović and R. Cipolla. An information-theoretic approach to face recognition from face motion manifolds. Image and Vision Computing, 24(6), 2006.
	O. Arandjelović and R. Cipolla. Incremental learning of temporally-coherent Gaussian mixture models. In Proc. British Machine Vision Conference, 2:759-768, 2005. [LINK]
	O. Arandjelović, G. Shakhnarovich, J. Fisher, R. Cipolla and T. Darrell. Face recognition with image sets using manifold density divergence. In Proc. IEEE Conference on Computer Vision Pattern Recognition, 1:581-588, 2005. [LINK]

Description

The database contains 100 individuals of varying age, ethnicity and gender, see a summary in Fig. 1. For each person in the database we collected 14 video sequences of the person in arbitrary motion. We used 7 different illumination configurations and acquired 2 sequences with each for a given person. The individuals were instructed to approach the camera and move freely, with the constraint of being able to see their eyes on the screen providing visual feedback in front of them, see Fig. 2. Most sequences contain significant yaw and pitch variation, some translatory motion and negligible roll. Mild facial expression changes are present in some sequences (e.g. when the user was smiling or talking to the person supervising the acquisition).

(a) Gender distribution
Male	67 individuals
Female	33 individuals

(b) Age distribution
18-25	30 individuals
26-35	45 individuals
36-45	14 individuals
46-55	7 individuals
56-65	4 individuals

Fig. 1. Database contents summary.

Fig. 2. The user is given visual feedback on the screen.

Download an example of a 10s sequence (794 KB): [LINK]

Acquisition hardware and miscellaneous

Video sequences were acquired using a simple pin-hole camera with automatic gain control, mounted at 1.2m above the ground and pointing upwards at 30 degrees to the horizontal, see Fig. 3. Data was acquired at 10fps, giving 100 frames for each 10s sequence, in 320 by 240 pixel resolution, see Fig. 4. On average, the face occupies an area of 60 by 60 pixels.

(a) Pin-hole camera used for data acquisition.

(b) Geometric configuration.

Fig. 3. Data acquisition setup.

Fig. 4. Frames from a typical video sequence.

Obtaining the database

The acquisition of data for the Cambridge Face Database was funded by the Toshiba Corporation. Therefore, at present, its content is not publicly available. However, if you are interested in using the database or a part of it, and believe that this could benefit the University of Cambridge or the Toshiba Corporation, please do contact me.