Research On Facial Expression Recognition

With the prominent application prospects and market value of emotionrecognition in driving, detection and game industry, facial expression which is the keycomponent of emotion recognition has become key challenge in the field ofanthropomorphic new human-computer interaction. In this paper, for the differenttypes of input pictures: single picture and picture sequences, complete static anddynamic expression recognition respectively.For single picture, an approach is presented for facial expression recognitionthrough the shape of facial feature points and texture information of specific areas,based on Active Appearance Model (AAM). First, find out that the shape and textureparameters can express more personalized information of each expression. And thenuse these two features to classify expressions, which rely on machine learningclassification algorithm. Next, weigh the relationship between identification rate andretaining information in the training process of AAM, and understand that noise isalso introduced into the classification along with the increase of retaining information.So it is necessary to get high identification rate in a lower dimension.For picture sequences, based on statistical analysis of the relationship betweenexpression and changes of characteristic organs, extract the features of expression,then select the subset of features through feature selection method. Finally, transformthe feature space to classify the expressions. Firstly, extract the features reflectingdifferences of various expressions and add the optimization features which makedifference more obvious. Then assess the features through Support Vector Machine tofurther streamline the commonality. Ultimately, to achieve expression recognition,feature space is projected onto the base vector space relying on least squares method.This paper extracts features more suited to facial expression recognition. Besidesreduce feature dimension effectively, under without loss of recognition rate conditions.Meanwhile, in order to realize the facial expression recognition, this paper projectsthe feature space onto base vector space relying on the least squares method. Researchperspective and methods of this paper optimize the efficiency and accurate of staticand dynamic expression recognition, at the same time, makes a contribution to voiceand movement combined with cross-modal emotion recognition.