IJCB 2011 Competition on counter measures to 2D facial spoofing attacks
Competition on counter measures to 2D facial spoofing attacks
International Joint Conference on Biometrics 2011
11th to 13th of October 2011
Washington DC, USA
Facial recognition techniques are currently deployed in a wide range of applications for access control, security, law-enforcement and database indexing, to cite a few.
As in any other field of Biometrics such as speaker or fingerprint verification, 2-D facial recognition systems are subject to attacks. In direct attacks, potential intruders try and may gain access to classified information by interacting directly with the system input cameras, like a normal user would. Such attempts are commonly referred as "spoofing".
For this competition we would like to measure the performance of algorithms that act as a counter-measure to spoofing. You are invited to submit your technique for evaluation.
For the performance measurement, a database of printed-photo attacks will be made available. It consists of 200 videos of printed-photo attack attempts to 50 clients, under different lighting conditions. The database also contains 100 reference enrollment and 200 real-access attempt videos. The objective is to measure the impact of your counter-measure or algorithm in avoiding attacks (false acceptance) and letting real accesses through (false rejection).
We provide are 3 video samples:
|Real Access||Hand-based Printed Photo Attack||Fixed-based Printed Photo Attack|
|Download file in||Download file in MOV||Download file in|
The database will be divided in three sets for training, development and testing containing respectively 30%, 30% and 40% of the total data. The organizers will distribute initially the training and development sets so competitors can go on building their counter-measures. Attacks and real-access will be annotated.
Towards the end of the period, organizers will make available the final testing set, on which results shall be evaluated. The testing data will not be annotated (correct label unknown).
After going through the testing data, competitors should send organizers their score files with scores annotated for each of the video sequences in a single text file containing the base filename of the video sample being tested and a score, between 0 (attack) and 1 (real-access). Example:
Organizers will provide a simple web-based interface to upload and retrieve results as PDF and/or PNG plots that can be included in publications. Please note that according to the IJCB'11 organization, the participants are free to submit individual papers describing their approaches. In addition, a single competition paper will be written describing and summarizing the techniques provided by competitiors.
How to engage
A Doodle poll is available to which competitors should subscribe in order to receive specific information. Competitors will have to agree and sign an End User License Agreement to gain access to the data.
- Distribution of "training" and "development" data sets: February, 1st 2011
- Distribution of "testing" set: April, 1st 2011
- Results to be submitted to the organizers by: May 3rd 2011
- Descriptions for the competition paper: from May 3rd to May 25 2011
Dr. Sébastien Marcel ( sebastien DOT marcel AT idiap DOT ch)