University of Twente Student Theses


Fusing Forensic Features and a Face RecognitionSystem on Lookalike Faces

Tijink, M.L. (2022) Fusing Forensic Features and a Face RecognitionSystem on Lookalike Faces.

[img] PDF
Abstract:Face Recognition Systems are popular and widelyused, however their performance on challenging cases can stillbe improved. One of the challenges are lookalikes, which aresubjects who look similar, but have a different identity. In thisresearch the dependence of the score computation of an existingFace Recognition System on face regions will be analyzed. Byoccluding parts of the face and visualizing the change in score asheatmaps, the cases of mated, (random) non-mated and lookalikepairs can be compared. The heatmaps show that the regions ofthe eye(brows) and nose are important for mated and lookalikepairs. The next step is to investigate whether the performanceof the Face Recognition System can be improved by fusing itsoutput with forensic features. The idea is to force the wholesystem to pay attention to details in important facial regions.A methodology is presented to automatically retrieve forensicfeatures from an image. Several fusing strategies are comparedwith a focus on the case of lookalikes. Results show that, althoughoverall performance is not significantly improved, comparableresults with a face recognition system are reached and severalfused systems show potential on individual lookalike cases.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:53 electrotechnology, 54 computer science
Programme:Electrical Engineering MSc (60353)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page