基于提升小波和FLD的人脸表情识别

Facial expression recognition based on lifting wavelet and FLD

提出一种基于提升小波和Fisher线性判别法(FLD)相结合的人脸表情特征提取方法。提升小波是完全基于时空域的变换,具有多分辨率的特征,更有利于表情细节信息的提取,并且运算时间短,便于实现。图像经过提升小波变换后,取其低频分量和高频分量相结合作为整体特征,实验证明保存了绝大部分的表情分量,然后用Fisher线性判别法(FLD)进行特征提取,采用K-近邻法进行分类。在JAFFE数据库中,分辨率达到94.3%,识别时间为2.9s,证明了方法的有效性。

A new facial expression feature extraction method based lifting wavelet and FLD is presented. The lifting wavelet is transformed completely in time-space domain and has the multi-resolution characters, so it is advantageous in dealing with feature extraction of the image＇s details. The result shows that the whole character made up by the LF and HF components contains the main expression feature of the expression image. Then the Fisher linear discriminant （FLD） is used to extract features from the lifting wavelet processing images. The K-neighbor method is used for classification. Experiment shows effective2y that in the JAFFE database recognition rate reaches 94. 3% and recognition time only lasts 2. 9s. The new methods proves to be faster and more effective.