Determining Facial Attributes From Geometric Features
Facial attributes are commonly determined via appearance type features combined with SVM and/or CNN. These data types have complex relations to the attributes and as such require complex methods. This paper presents a simpler approach using geometric methods. Geometric features are derived from landmarks about the face into a variety of different types, from triangle areas to distances between points. These are then given directly to Random Forest (ExtraTrees) and multilayer perceptron classifiers (MLP). Further, a very small test is done on the challenging SCface dataset, where the MLP classifier ...
(For more, see "View full record.")