ICBA2006 Face Verification Contest on the Xm2vts dataset

Organised by Kieron Messer and Josef Kittler
University of Surrey, Guildford, Surrey. GU2 7XH, UK.

Overview

In order to promote evaluation of pattern recognition algorithms using publicly available datasets and standard performance assessment methodology, a competition for the best face verification algorithms will take place in conjunction with the ICBA 2006 conference.

The authors of the best performing algorithm of the competition will receive a financial prize that will be presented at the Conference Dinner. The performance of all submitted algorithms will be summarised in a presentation given at the conference.

The tests will be carried out using images acquired from the Xm2vts database on the standard and darkened image sets. Refer to ‘Purchase the database’ in the left hand frame. The last competition to be run on the Xm2vts database was held in conjunction with AVBPA 2003 and the results of which were published in this paper.

Part I: Standard Test

The first tests should be carried out on images taken from CDS001 and CDS006 of the Xm2vts database. Results need to be submitted using your algorithms on Configurations I and II of the Lausanne protocol. This includes the Equal Error Rate on the evaluation data set and the FA and FR rates on the test data set. We will also require you to submit your scores for each of the image tests. The files below will help you generate the correct sets of data for the competition.

• Manual Eye Positions
• Config I - Evaluation
• Config I - Test
• Config II - Evaluation
• Config II - Test

Part II: Darkened Image Test

This test is designed to test the system robustness to varying lighting conditions. It will be carried out on images taken from CDS008 of the Xm2vts database. This dataset contains images of the same 295 Xm2vts subjects but in varying illumination conditions. In this test the same enrolments as for Configuration I should be used. In the test phase of the protocol the 4 images from the corresponding dark image set should be used instead of the normal test images. You will be required to submit your scores, FA and FR results on this test data. The files below will help you generate the correct sets of data for the competition.

• Manual Eye Positions
• Dark Image Test

Results Format

To submit results to the competition you are required to provide the following information.

• Which part of the competition the results refere to.
• A short description of the method used. This should be no more than 250 words.
• Bibtex reference to a more complete description of your method.
• Whether automatic localisation or manual co-ordinates has been used.
• For Part I the EER on the evaluation dataset.
• For Part I the score file for the evaluation set.
• For Parts I and II the FAR and FRR on the test set.
• For Parts I and II the score file for the test set.

The score files should have the following format

	actualId claimedId imageId score

For example, the first few lines of an Xm2vts-Config-I evaluation file might look like.

	000 003 000_1_1 21.12
	000 003 000_1_2 12.34
	000 003 000_2_1 -19.03
	.........

You can enter each part of the competition as many times as you like.

Important Deadlines

• Results submission September 17th 2006

For any further information please contact Kieron Messer (k.messer@surrey.ac.uk).

---