SPATIALLY FNMF-WEIGHTED LOCAL STEERABLE FEATURE FOR FACIAL EXPRESSION RECOGNITION
Feature extraction, selection and representation are critical steps for facial expression recognition. The approach proposed in this study generalizes and combines the above typical issues in a unified way. Firstly, a set of steerable filters which have been successfully used in various computer vision applications are used to extract local feature, which is locally stable and distinctive enough to capture subtle facial expression cues. Then, a spatial weight function is calculated based on the Fisher-NMF method to quantify the spatial distribution of expression information in different facial areas. Based on the weight function, local feature in facial areas containing abundant expression information is selected to represent expression patterns in face images. The weight information is also embedded into the classifiers for the final decision. Experiment results on the JAFFE Database demonstrate the effectiveness of the proposed method.
Facial expression recognition Steerable filter NMF LNMF Fisher-NMF (FNMF)
SHUANG XU YUNDE JIA LEI SHI YOUDONG ZHAO
Beijing Institute of Technology, Beijing 100081, P R. China
国际会议
开封
英文
457-467
2006-10-15（万方平台首次上网日期，不代表论文的发表时间）