DNA analysis techniques could dramatically improve facial recognition accuracy
The techniques of genomic analysis could be applied to facial recognition to substantially increase accuracy, according to an article by Associate Professor of Pattern Recognition at Kingston University Jean-Christophe Nebel in The Conversation.
Nebel sights the use of facial recognition by UK police at the Notting Hill Carnival in August, which resulted in approximately 35 false matches and one “erroneous” arrest, as typical of the accuracy limitations of current facial recognition technology when dealing with “real-life situations.”
The effectiveness of automated video analysis software could be dramatically improved, however, by the application of software innovations developed for DNA sequence analysis, according to Nebel. The large amounts of variable data involved in identifying individuals in public settings is in contrast to the relatively static environments in which the accuracy of object and facial recognition have been demonstrated.
“In order to improve automated video analysis, we need software that can deal with this variability rather than treating it as an inconvenience – a fundamental change,” Nebel writes. “And one area that is used to dealing with large amounts of very variable data is genomics.”
The money and resources dedicated to dealing with variable data for genomic analysis has supported the development of software and computer resources to treat differences as mutations, a technique which can be applied to video to track moving individuals.
“Early tests of this “vide-omics” principle have already demonstrated its potential,” Nebel states. “My research group at Kingston University has, for the first time, shown that videos could be analysed even when captured by a freely moving camera. By identifying camera motion as mutations, they can be compensated so that a scene appears as if filmed by a fixed camera.”
University of Verona researchers have demonstrated an approach which may make software development much faster and less expensive, as standard genomic tools can be applied to video with image processing tasks encoded a particular way.
Ultimately, Nebel predicts that “video-omics” could drastically change mass facial recognition in uncontrolled environments within a decade.