Publication Details
Overview
 
 
Isabel Gonzalez, , Dongmei Jiang, Werner Verhelst, Hichem Sahli
 

Chapter in Book/ Report/ Conference proceeding

Abstract 

We present a framework for combination aware AUintensity recognition. It includes a feature extraction approachthat can handle small head movements which does not requireface alignment. A three layered structure is used for the AUclassification. The first layer is dedicated to independent AU recognition, and the second layer incorporates AU combinationknowledge. At a third layer, AU dynamics are handled based onvariable duration semi-Markov model. The first two layers aremodeled using extreme learning machines (ELMs). ELMs haveequal performance to support vector machines but are computationallymore efficient, and can handle multi-class classificationdirectly. Moreover, they include feature selection via manifoldregularization. We show that the proposed layered classificationscheme can improve results by considering AU combinations aswell as intensity recognition.

Reference