基于贝叶斯判别器的面部检测

 提出了1种基于PCA(主成分分析)的贝叶斯判别器用于检测灰度面部图像.为检测面部图像,首先用PCA减低训练图像的维数以为判别器提供教好的图像描述.训练图像包括面部图像和非面部图像并给出正确标识,用EM算法学习图像的特征向量.在构建好学习模型后,用贝叶斯后验概率检测未知样本.模型参数估计和判别原则都是基于最大似然度.在估计了概率密度函数后,贝叶斯判别器可产生最小的误差,为分类的教优准则.本方法用2356副面部图像和3780非面部图像作为学习样本,学习过程获取面部图像与非面部图像的差异而构建判别模型.训练图像包括不同位置,不同表情,不同亮度条件的同一对象图像.训练模型用于检测205副面部图像,实验结果在文章第4部分给出.

Face detection with Bayesian classifier

A method using Bayesian classifier is presented,which is based on PCA(Principle Component Analysis) for face detection in gray scale images.The method first applies Principle Component Analysis(PCA) to reduce the dimensionality of train images and provide better resulting image representation for classifier task.We first train a model with labeled images,which are face images and non-face images based on eigenvectors with EM algorithm.After building the model a test face image may be detected by Bayesian posterior probability.The parameters of the model are estimated with maximum likelihood and the discrimination rule is also based on maximum likelihood.The Bayesian classifier yields the minimum error when the underlying probability density function is known.It’s the optimal criterion for classification because it measures class separately.We use 2?356 face images and 3?780 non-face images as train images to capture the variation of human faces and the difference between human face and non-human face.Train face images include ones in different lighting poses,with different expressions and under different lighting conditions.The method has been used to detect 205 images and shows good performance.The experimental result is displayed.