基于小波和ICA的面部特征定位

面部特征的定位是自动人脸识别（AFR）系统的重要组成部分，现有主要包括基于先验知识、几何形状、色彩、外观和关联信息五类方法。本文从信号学的角度，提出了一种基于小波和独立分量分析（ICA）的新方法。先对面部图像进行小波分解，提取出主要代表眼睛和嘴巴特征的水平边缘图像，再把说话人在视频流中眼睛的闭合和嘴巴的运动看成是相互独立的运动分量，利用ICA分离出眼基和嘴基，然后分别利用它们来重建人脸图像，从而实现眼睛和嘴巴的定位。

Facial Feature Localization Based on Wavelet and ICA

Facial feature localization, an important technique in automatic face recognition, could be classified into five methods, namely ones based on knowledge, geometry information, color, appearance and relative location. Our new method based on wavelet and independent component analysis regards the movements of eyes and mouth as independent components from the view of the signal processing. The wavelet decomposition is first applied to obtain the horizontal edge images. The apparatus basis images, extracted by ICA on these edge images, are then used separately to recon struct the facial image to localize eyes and mouth.