Quality-based ROC calibration for forensic face recognition

Description of research

Forensic investigators now have access to video recording of almost every crime scene; thanks to omnipresent CCTV camera. The face image extracted from such CCTV footage is of very low quality (in terms of pose, illumination, imaging noise, etc.).

It is know that even the current state-of-the-art face recognition system is known to have very poor recognition performance on low quality CCTV face images. Therefore, instead of using automatic face recognition systems, investigators working in forensic cases prefer to use human experts trained in manual forensic face recognition.

Performance of a face recognition system is known to change with the quality of face images. My PhD research aims to explore a relationship between deviation in face image quality and corresponding variation in recognition performance. Mathematical model of such a relationship can be used to improve the evidential weight of recognition results in a forensic face recognition case.

Advisor(s)

Dr.ir. R.N.J. (Raymond) Veldhuis

Dr.ir. L.J. (Luuk) Spreeuwers

Duration

December 31, 2010 to December 31, 2014

Project

Bayesian Biometrics for Forensics - BBfor2

Funding institution

BBfor2 (BBfor2 is funded by the EC as a Marie-Curie ITN-project (FP7-PEOPLE-ITN-2008) under Grant Agreement number 238803).

Department of Electrical Engineering Mathematics and Computer Science (EEMCS), University of Twente.

Strategic Research Orientation

IE&ICT - Industrial Engineering & ICT

Links to relevant web pages

http://abhishekdutta.org/phd-research/

http://bbfor2.net

Pictures

	CTIT Home > Meet the PhDs > Abhishek Dutta
Home CTIT symposium 2013 About CTIT News NIRICT Research Meet the PhDs Working at CTIT Internal Information Conferences Open Calls (updated May 3, 2013) Library and Videos Knowledge Transfer Links Sitemap Search	Quality-based ROC calibration for forensic face recognition Description of research Forensic investigators now have access to video recording of almost every crime scene; thanks to omnipresent CCTV camera. The face image extracted from such CCTV footage is of very low quality (in terms of pose, illumination, imaging noise, etc.). It is know that even the current state-of-the-art face recognition system is known to have very poor recognition performance on low quality CCTV face images. Therefore, instead of using automatic face recognition systems, investigators working in forensic cases prefer to use human experts trained in manual forensic face recognition. Performance of a face recognition system is known to change with the quality of face images. My PhD research aims to explore a relationship between deviation in face image quality and corresponding variation in recognition performance. Mathematical model of such a relationship can be used to improve the evidential weight of recognition results in a forensic face recognition case. Advisor(s) Dr.ir. R.N.J. (Raymond) Veldhuis Dr.ir. L.J. (Luuk) Spreeuwers Duration December 31, 2010 to December 31, 2014 Project Bayesian Biometrics for Forensics - BBfor2 Funding institution BBfor2 (BBfor2 is funded by the EC as a Marie-Curie ITN-project (FP7-PEOPLE-ITN-2008) under Grant Agreement number 238803). Department of Electrical Engineering Mathematics and Computer Science (EEMCS), University of Twente. Strategic Research Orientation IE&ICT - Industrial Engineering & ICT Links to relevant web pages http://abhishekdutta.org/phd-research/ http://bbfor2.net Pictures