一种新的多人脸检测方法研究

A New Method Research of Multi-Face Detection

自动人脸检测是人脸识别系统的一个重要部分,本文提出了一种新的基于独立成分分析(ICA)和多项式神经网络(PNN)相结合的人脸检测方法,该方法首先在训练样本中使用ICA分离出代表人脸和非人脸特征子空间的独立影像基,把训练图像映射到该子空间降维后作为PNN网络的输入训练网络;对测试图像采用移动多尺度窗口提取图像模式,采用ICA降维后输入PNN网络,进而分类检测出人脸和非人脸。算法通过CMU-MIT的复杂背景人脸库中的多人脸图像进行实验,得到很高的检测率和较低的误检率.

Automatic face detection is an important component for face recognition system. This paper proposed a new face detection method based on independent component analysis（ICA）and polynomial neural network （PNN）. Firstly this method using ICA separates the image base standing for the feature subspace of face and non-face in the training samples. After projecting the sample images to this subspace, we get the reduction features and use them as the input to train the singer layer PNN. To the test images, using mul- ti-scale moving windows to extract image patterns, then after using ICA reduction, input them to PNN to classify them as face or non-face . In experiments on CMU-MIT image base with complex backgrounds, the proposed method has produced high detection rate and low false positive rate.