摘要
研究了将人脸作为一种特殊的图像内容进行检索的问题 ;采用基于Adaboost统计学习方法的层叠分类器检测人脸 ,再用非线性SVM分类器验证人脸 ;实现了在大规模的复杂背景图片集合中高速准确的人脸定位 ;为了将找到的人脸规范化 ,借鉴直接表观模型 (directappearancemodel) ,提出了一种新的特征检测和人脸校正方法 .该方法基于对大量数据的统计学习过程 ,具有良好的扩展性和稳定性 ;在此基础上 ,采用SVM分类器实现了人脸检索 ;最后 。
In this paper, the problem of face retrieval is discussed in which faces are treated as a particular type of content in images. A high speed accurate face localization system is implemented for large scale image database with complex background, in which the cascade classifier based on Adaboost statistical learning method is used for face detection and a non linear SVM classifier is used as a face verifier. For face normalization, a new feature detection and calibration method based on Direct Appearance Model is developed, which is very adaptive and robust. Then face retrieval is implemented based on SVM classifier. Finally experiment results are given to demonstrate the effectiveness of the above method.
出处
《计算机学报》
EI
CSCD
北大核心
2003年第7期874-881,共8页
Chinese Journal of Computers
基金
国家自然科学基金项目 ( 60 2 73 0 0 5 )资助
关键词
人脸检测
人脸检索
支持向量机
直接表现模型
face detection
face retrieval
adaboost
support vector machine
direct appearance model
