A Coarse-to-Fine Classification Scheme for Facial Expression Recognition

Abstract. In this paper, a coarse-to-fine classification scheme is used to recognize facial expressions (angry, disgust, fear, happiness, neutral, sadness and surprise) of novel expressers from static images. In the coarse stage, the sevenclass problem is reduced to a two-class one as follows: First, seven model vectors are produced, corresponding to the seven basic facial expressions. Then, distances from each model vector to the feature vector of a testing sample are calculated. Finally, two of the seven basic expression classes are selected as the testing sample’s expression candidates (candidate pair). In the fine classification stage, a K-nearest neighbor classifier fulfils final classification. Experimental results on the JAFFE database demonstrate an average recognition rate of 77% for novel expressers, which outperforms the reported results on the same database.

• Type: Conference
• Year: 2004
• Where: ICIAR
• Authors: Xiaoyi Feng, Abdenour Hadid, Matti Pietikäinen
• Keywords: Facial Expressions, ICIAR 2004, Model Vector, Testing Sample