Abstract:

Micro-expression,as an important channel to reveal underlying emotion,is a kind of unconscious nonverbal facial information,not controlled by the brain,can reflect people's real psychological experience and psychological state,but micro-expression movements appear small and quick,and cannot be easily captured,making it difficult for a single mode of micro expression recognition accuracy to ascend.To solve the above problems,this paper proposes an algorithm for extracting the facial color features of micro-expressions,and integrates the extracted features with the texture features of micro-expressions for decision fusion,so as to construct the bi-modal emotion recognition model of micro-expressions.First,the model extracts the corresponding texture features from the preprocessed micro-expression data by the uniform LBP-TOP algorithm.Second,the Lab color difference between each pixel of two frames of micro-expression sequence images is calculated to obtain the facial color features,and the embedded feature selection is carried out to eliminate redundant features.Then,the classifiers of the two modes are trained respectively,and the classification information obtained from the two modes is fused for decision making.Finally,the classification results of micro-expressions are obtained.The model was tested on CAMSE Ⅱ and SMIC micro-expression dataset.Experimental results show that the average recognition accuracies of the micro-expression single mode of texture and face color are 64.73% and 51.64%,and 63.58% and 50.48%,while the results of micro-expression emotion recognition after decision fusion are 68.11% and 66.43%.The recognition accuracy is higher than that before the fusion,which indicates that the proposed bimodal emotion recognition model can significantly improve the recognition ability of micro expressions.

Key words: facial color feature, texture features, euler video amplification, feature selection, decision fusion