基于高斯隐马尔可夫模型的人脸特征标注和识别
投稿时间:2019-06-17  修订日期:2019-06-25  点此下载全文
引用本文:
摘要点击次数: 134
全文下载次数: 0
作者单位邮编
安晓宁* 广西科技大学 000000
王智文 广西科技大学 
庚佳颖 广西科技大学 
李秋玲 广西科技大学 
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目)
中文摘要:传统的基于隐马尔可夫模型的人脸识别方法需要对原始人脸图像进行光照补偿、人脸旋转等预处理,而且模型对人脸姿势、表情、局部特征变化等非常敏感。文章提出一种基于高斯隐马尔可夫模型的人脸特征标注方法,该方法假定人脸图像中人脸和人脸特征两个区域的灰度值服从两个不同的高斯分布,并将这两个分布作为隐马尔可夫模型的状态集合。该方法将灰度人脸图像转换为一维的灰度值序列作为观测序,通过模型预测状态序列以实现人脸特征的标注和定位,并基于该模型建立人脸数据库,对未知人脸进行识别。该方法在ORL人脸库和自建人脸库测试中,均取得较高的标注准确率和识别准确率。
中文关键词:高斯隐马尔可夫模型,特征标注,人脸识别,ORL人脸库,自建人脸库
 
Face feature labeling and recognition based on Gaussian Hidden Markov Model
Abstract:The Traditional face recognition method based on Hidden Markov Model needs to preprocess the original face image, such as illumination compensation and face rotation, etc., and the model is sensitive to face pose, expression, local feature changes and so on.This paper proposes a face feature labeling method based on Gaussian Hidden Markov Model, which assumes that the gray values of face and face feature in face image obey two different gaussian distributions, and takes these two distributions as the state sets of Hidden Markov Model. The method converts gray face image into one-dimensional gray value sequence as the observation sequence, achieves the annotation and positioning of facial features by predicting the state sequence of the model, and establishes a face database based on the model to recognize unknown faces. This method has achieved high annotation accuracy and recognition accuracy in ORL face database.
keywords:hidden markov model  feature labeling  face recognition  ORL face database  self-built face database
查看全文   查看/发表评论   下载pdf阅读器