利用几何特性及神经网络进行人脸探测技术的研究

在人脸识别过程中 ,首先也是最重要的一个环节是人脸探测 ,因为一旦从图像中定位并提取到了人脸 ,那么下一步的人脸识别工作就变得非常容易。眼睛是人脸图像中最容易探测的部位 ,而且通过探测双眼来发现人脸最符合人的视觉习惯。提出了一种基于几何特征分析和人工神经网络的由粗到细的两级人脸探测方法。在第一级中 ,眼睛和脸是通过测量眼睛的尺寸和眼睛与脸的位置关系探测到的 ,第一级的输出是一个尺寸归一化的人脸 ,但偶尔也伴随着一个或多个因对复杂背景中与眼睛类似的物体的误判而得到的非人脸图像 ;第二级神经网络正是用来过滤掉第一级中被误判的人脸。实验表明 ,这种由粗到细的两级人脸探测系统具有很高的稳定性和探测正确率

Face detection method based on structural analysis and neural network

Face detection is the most important prior process to face recognition, for once a face is located exactly in any scene the face recognition would become easy There are many facial structural features to describe a face The eyes are the easiest parts to be detected and detecting the eyes to find a face is coherent with the visual habit of human beings In this paper a novel approach concerning a coarse to fine face detecting method is proposed that employed face detecting system with two orders based on geometric analysis and neural network In the first order, eyes and face are detected by means of their geometric property such as the dimensions of the eyes and the relations between eyes and face The output of the first order is a normalized face but occasionally companied with one or more none face caused by the misanalysis of the eyes like objects in a complex background The second order is a well trained neural network, which is used to filtrate the misanalyzed outputs in the first order This face detecting system with two orders is proved to be robust and has high accurate detecting rate