
Der neue Algorithmus vergleicht Hunderte von 3D -Gesichtern tausendmal schneller als alte Methoden, indem ein berechnete „durchschnittliche Gesicht“ als clevere Verknüpfung verwendet wird und in Minuten anstelle von Tagen nahezu die gleiche Genauigkeit erreicht
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0329489
2 Kommentare
Using an „average face as a shortcut“ is a well known technique for detecting 2D faces. The technique is called „eigenfaces“, or more generally „principle component analysis“. It works by calculating the most dramatic ways that faces differ from the average one (the „principle components“ with large „eigenvalues“), and then analyses faces based on only these important ways, disregarding the less-important ways, saving processing effort.
This is great news until it gets used in surveillance technology and we have nowhere else to hide. It’s all fun and games till we have no choice over it’s usage.