摘要
文章所述头部检测方法,首先采用Mallat小波分解提取的头部特征,作为神经网络的输入,通过构造训练集对BP网络进行训练,自举训练方法提高了神经网络的泛化能力。在此基础上,通过在整个图象上自左至右、自上至下移动窗口,用训练好的神经网络对待识模式进行识别,最后根据启发式规则进行后处理。该算法已经应用于头部跟踪项目中,取得了较好的效果。
An application of human's head detection using wavelet transform and artificial neural network is discussed in this paper.Comparing with face detection,head top detection utilizes less features with less distinctions.The system arbitrates with artificial neural network to detect a head top.Wavelet transform acts as an approach to extract informtion of high and low frequency from the pattern,while bootstrap is used to lift up the accuracy.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第4期86-88,101,共4页
Computer Engineering and Applications
基金
广州市科技计划项目"人运动跟踪识别预警系统"的资助(编号:2002XGP16)
广州市天河区科技局科技计划项目"基于动态人脸识别的身份认证系统"的资助(编号:2002XGP11)
