摘要
提出了一种基于Gabor小波变换和两次离散余弦变换(DCT)相结合的人脸表情特征提取方法,在保留有效的纹理信息基础上降低了表情特征的维数.首先对人脸表情图像进行第一次DCT压缩图像,然后对处理过的图像执行Gabor变换,提取表情特征,进而对得到的不同尺度和方向的特征图像进行第二次DCT,得到包含大量表情信息的低维特征向量,最后用BP神经网络对特征向量分类.实验结果显示该方法识别率较高.
The Gabor transform is combined with two times discrete cosine transform（DCT）to extract facial expressional features.Effective texture information is retained while reducing the dimension of expression characteristics.First compress the facial expression image with DCT.Second implement Gabor transform to image,to excavate expression features,then implement the second times DCT to the different scale and orientation＇s characteristics images that get by Gabor transform,to attain low-dimensional features vector which includes a large number of expression information.Finally,classify the features vector with BP neural network.The experiment result shows that this method gets a higher recognition rate.
出处
《微电子学与计算机》
CSCD
北大核心
2009年第5期262-264,共3页
Microelectronics & Computer
关键词
表情识别
GABOR变换
离散余弦变换
神经网络
facial expression recognition
Gabor
discrete cosine transform（DCT）
neural network
