Real-world face recognition requires an ability to perceive the uniqueness of a face across multiple, variable images. Deep convolutional neural networks (DCNNs) accomplish this feat and can be analyzed in a multidimensional “face space”.
We examined the organization of viewpoint, illumination, gender, and identity in this space. Specifically, we probed a DCNN trained with in-the-wild images with an in-the-lab dataset consisting of rendered images from 3D laser scans of faces. We show that DCNNs create a highly organized face similarity structure in which identities and images coexist. Natural image variation is organized in this hierarchy, with face identity nested under gender, illumination under identity, and viewpoint under illumination. To examine identity, we caricatured faces and found that identification accuracy increased with caricaturing. Mimicking human perception, DCNN caricature representations “resembled” their veridical counterparts. Caricaturing also minimized illumination and viewpoint representational variation. DCNNs offer a theoretical framework that reconciles decades of behavioral and neural results that emphasized either the image or the object/face in representations, without understanding how a neural code could seamlessly accommodate both.
Alice J O'Toole is a Professor at the University of Texas at Dallas in the School of Behavioral and Brain Sciences. Her research interests include human perception, memory and cognition. In 2007, she was named Aage and Margareta Møller Endowed Professor. She currently serves as an Associate Editor of Psychological Science and the British Journal of Psychology and has served as Program Chair of the 2017 IEEE Meeting on Automatic Face and Gesture Recognition. Alice received a BA in Psychology (1983) from The Catholic University of America, Washington, DC, and a MS (1985) and PhD (1988) in Experimental Psychology from Brown University, Providence, RI. Dr. O’Toole has been the recipient of an Alexander von Humboldt Research Fellowship, and has received research funding from NIH, NIMH, DARPA, IARPA, and the National Institute of Justice.
Distinguished Lectures Series in Cybersecurity
Mit der Distinguished Lecture Series in Cybersecurity kommen jedes Semester herausragende Experten aus Wissenschaft und Wirtschaft nach Darmstadt, um die vielfältigen Chancen und Herausforderungen im Bereich der IT-Sicherheit zu diskutieren.
Die Redner stellen in den Vorlesungen richtungsweisende Forschungsergebnisse verschiedener Disziplinen vor, fassen komplexe Themenbereiche zusammen und zeigen den momentanen Kenntnisstand ihres Forschungs- oder Arbeitsgebietes auf.
Die Vorträge sind öffentlich. Interessierte Besucher sind herzlich willkommen. Nach der Vorlesung gibt es Möglichkeit zum Austausch bei einem kleinen Buffet vor Ort.