Acquisition of a large database for biometric identity
veri?cation
K.Messer,J.Matas and J.Kittler
University of Surrey
Guildford Surrey GU25XH United Kingdom
Summary
it to the centre.They were told that a full side-pro?le was required and asked to run through the entire sequence twice.Figures1(a)-(h)show selected frames from this sequence.This sequence was kept constant for all four sessions.Figure2(a)-(d)show a front pro?le image taken for a subject taken at each of the four sessions.Figure2(e)-(h)shows selected frames taken from one of the speech shots.
Next,if the subject was wearing glasses he/she was asked to remove them and a short front pro?le video sequence was?lmed.In total about1.5minutes of digital video was taken per sub-ject,per session.
During the third session a3D model of each subjects head was acquired using a high precision stereo-based3D camera developed by the Turing Institute[4].
The entire database was acquired using a Sony VX1000E digital cam-corder and DHR1000UX digital VCR.This captures video at a colour sampling resolution of4:2:0and audio at frequency 32kHz.This hardware was chosen because it can be interfaced to a computer via a?rewire port. Software has now been written that allows a user to capture video sequences,audio sequences and individual frames to the computer hard disk directly from the VCR.
The full distribution of the database will come on approximately161-hour Mini-DV cassettes. Digital video is compressed at the?xed ratio of5:1so the entire database occupies a data volume of0.4Tbytes(2Tbytes uncompressed).In order to use the database in this form the user will re-quire a digital video recorder/camera and a computer with appropriate software.At the time of writing the only available option is a PC with?rewire card but other manufacturers such as Sun and Silicon Graphics have made it known that support for?rewire will soon be made available in their systems.
However,for ease of use,pre-selected image,audio and video subsets of the database intend to be made available on CDROM,linear tape and IOMEGA Jaz drives.The reader is pointed to https://www.wendangku.net/doc/9a18313501.html,/Research/VSSP/xm2fdb to?nd out about which subsets are currently being made available.
Although this large collection of speech,video and3D data was commissioned in connection with biometric veri?cation,many other uses are envisaged,eg.training of lip-tracking and lip-reading systems,face detection and animation.
Conclusions
[1]Biometric technology today,ISSN0969-4765
[2]M.Acheroy,C.Beumier,J.Big—n,G.Chollet,B.Duc,S.Fischer,D.Genoud,P.Lock-wood,G.Maitre,S.Pigeon,I.Pitas,K.Sobottka,L.Vandendorpe Multi-modal person veri?ca-tion tools using speech and images.In the proceedings of the European Conference on Multime-dia Applications,Services and Techniques(ECMAST96);Louvain-la-Neuve,28-30May,1996.
[3]Devijver P.,Kittler J.,Pattern Recognition:A Statistical Approach,Prentice Hall,1982,
[4]https://www.wendangku.net/doc/9a18313501.html,/
(a)First centre shot (b)Head left at 45de-grees (c)Left pro?le shot (d)Head right at 45de-grees
(e)Right pro?le shot (f)Looking Up (g)Looking Down (h)Second centre shot Figure 1:Selected images from the rotating head sequence.The subject rotates head from left,to right,to up,to down then back to centre.
(a)Session 1(b)Session 2(c)Session 3(d)Session 4
(e)Speech 1(f)Speech 2(g)Speech 3(h)Speech 4
Figure 2:(a)-(d):Selected front pro?le shots taken of a subject from each of the four sessions.(e)-(h):Selected images taken from one of the speech shots.