提出一种基于小波分解和优选的VLBP特征的表情识别方法。该方法首先通过小波分解将原始图像分为几个不同频率的子图像来增强图像信息,然后用VLBP算子对不同频率的子图像运用不同的分块大小提取特征,采用神经网络贡献分析对特征进行选择,最后用SVM分类器进行识别。实验表明,该方法比单纯从原图像中提取VLBP特征更加有效,识别率更高,并且VLBP特征的提取速度快,可用于实时的人脸表情识别。
This paper proposed a new method of expression recognition based on wavelet decomposition and selected VLBP feature. This method enhanced original data by wavelet decomposition first, obtained several different frequency sub-images. Then extracted the features from these sub-images with different block size using VLBP operator. After that, selected features by contribution analysis algorithm of neural network. At last, recognized six expressions by support vector machine classifier. The experiments show that this method have higher recognition rate than pure VLBP feature extracted from original image. And tihs method can be used for real-time facial expression recognition because of its high speed.