基于混合特征和分层最近邻法的人脸表情识别

Human Facial Expression Recognition Based on Mixed Features and Hierarchical Nearest Neighbor Method

采用主动形状模型提取人脸嘴巴几何特征,利用Gabor小波提取眼睛和眉毛频域特征。根据人脸表情特征基于分类树思想将表情进行三层分类。第1层以嘴宽高比、嘴高、嘴宽高差作为最近邻的输入进行训练实现粗分类;第2层以嘴宽、嘴宽高差作为最近邻的输入进行训练实现分类;第3层以眼睛和眉毛区域15个关键点的Gabor小波特征作为最近邻的输入进行训练实现细致的分类。整个识别过程由粗到细,融合了几何特征和频域特征。实验结果表明该方法是有效的。

Geometry feature of the mouth is extracted by using Active Shape Model（ASM）, and frequency domain feature of the eye and eyebrow is extracted by using Gabor wavelet. According to characteristics of human facial expression, three-layer classification is proposed based on classification tree. The width, the ratio and the difference of width and height of mouth are taken as input of Nearest Neighbor（NN） to train and classify at the first layer. The width and the difference of width and height of mouth are taken as input of NN to train and classify at the second layer. The Gabor wavelet feature of 15 key dots of the eye and eyebrow are taken as input of NN to train and classify at the third layer. The whole identification process is from coarse to fine and combines geometric features and frequency domain features. Experimental results show that the approach is effective.