2020-10-08 on Computational workload reduction in biometrics

da/sec scientific talk on Computational workload reduction in biometrics

Topic: Accelerating facial identification with morphing

by Fabian Stockhardt
online Big Blue Button Room: D19/2.03a, 12.00 noon, October 08, 2020 (Thursday), 12.00 noon

Keywords — biometrics, workload reduction, morphing, identification, face recognition

Abstract

Despite steadily increasing computational power, increasing numbers of subjects in data sets can cause identification systems such as face recognition to take a unreasonable long time to produce results. This is mainly due to the fact that the systems have to compare the input subject to every person contained in the data set in order to perform a comprehensive identification. The morph acceleration method is able to reduce the total number of comparisons by fusing subjects at the sample level. Previous attempts have shown that it is possible to reduce the number of comparisons, but the identification rate is also reduced. By creating morph pairs based on similarity, the presented approach enables the system to considerably reduce the number of comparisons, without a loss in identificaiton rate.