Facial Expression Recognition Based On Active Appearance Models

Facial expression is the most important vector of human emotion, which containsa wealth of psychological and emotional information, and facial expression is the mostimportant non-verbal communication method. Recently, with the increase of people'sinterest in the human-computer interaction, facial expression recognition has becomeresearch focus and has a wide range of applications and potential market value.Generally,facialrecognitionsystemconsistsofmainlythreeparts,whicharefarcedetection, expression feature extraction and classi?cation. This paper presents a de-tailed description of the research on the key issues of expression feature extraction andclassi?cation. The main work of this paper is as follows:1.In the"expression feature extraction"part, we introduced Active AppearanceModels(AAM)into facial expression. AAM method is used to locate the position ofthe 68 points in the image sequences, then the di?erences of coordinates between theexpression peak frame and the neutral frame were computed as the expression feature.2.In the"classi?cation"part, we presented a new three-layer model. Our modelconsists of three primary layers, namely, feature layer, AU layer, and expression layer.From?rstlayertosecondlayerweconstructaSVMclassi?erforeachAU.Fromsecondlayer to third layer,we propose to use a Bayesian network(BN) to model the statisticaldependencies among AUs and expression.3.We introduced some methods about face detection and expression image pre-processing. And We conducted extensive experiments using CK+ database to validatethe method we proposed in this thesis.