摘要
人脸表情识别是一项充满挑战的工作,提出一种基于局部Gabor二值模式(LGBP)特征和稀疏表示的表情识别方法。对表情图像进行归一化处理,标定眉毛、眼睛、嘴巴等部位的特征点,划分出5个表情子区域。对各个子区域进行多尺度多方向的Gabor滤波,对Gabor系数图谱进行局部二值模式(LBP)编码,通过直方图方法降维,形成显著的特征向量。根据特征向量构建符合视觉特征的过完备字典,运用稀疏表示分类方法进行表情识别。通过在JAFFE表情库上进行实验,表情识别率达到87.5%,表明了该方法的有效性。
With the challenge of facial expression recognition, a novel method based on local Gabor binary pattern （LGBP） features and sparse representation is proposed. Firstly, the facial image is normalized, then set feature points in eyebrows, eyes and mouth and produce five local expression regions. Secondly, the loc-al regions are coded using a multi-orientation, multi-resolu- tion set of Gabor filters, then Gabor coefficient maps are coded using local binary pattern （LBP）, at last the facial expression image is modeled as a ＂histogram sequence＂ by concatenating the histograms of the local Gabor binary pattern maps of all the lo- cal regions. Finally, the over-complete dictionary is built and the sparse representation classification is using to recognize facial expression. The algorithm is experimented in Japanese female facial expression database （JAFFE database）, a recognition rate of 87.5 % is obtained and shows the effectiveness of the proposed algorithm.
出处
《计算机工程与设计》
CSCD
北大核心
2013年第5期1787-1791,共5页
Computer Engineering and Design
基金
河北省高等学校科学技术研究基金项目(Z2011293)
