摘要
提出一种基于DCT-BP神经网络的人脸表情识别算法,先对图像进行灰度均衡与图像平滑的预处理,然后利用离散余弦变换提取图像的表情特征参数,变换后的数据量大大减小,而且不会丢失图像所携带的关键信息,最后利用前向反馈神经网络算法进行识别。
An effective method for facial expression recognition is proposed, which uses the discrete cosine transform （DCT） and error back propagation （BP） neutral network. First, processing including normalization and filtering is carried out on facial images. Then two-dimensional discrete cosine transform （2-D DCT） is introduced. After that, the size of expression feature data is reduced without sacrificing key attributes that play important role in the recognition process. Finally, classification is made based on the back propagation （BP） algorithm.
出处
《微计算机信息》
北大核心
2005年第10S期142-144,共3页
Control & Automation
基金
上海市科技发展基金项目(7A17746)
关键词
表情识别
离散余弦变换
误差向传播算法
前向神经网络
facial expression recognition, discrete cosine transform, error back propagation algorithm, feedforward neural network
