动态权值预划分实值Adaboost人脸检测算法  

Dynamic Weights and Pre-partitioning Real-Adaboost Face Detection Algorithm

作　　者：武妍[1] 项恩宁[1] 

机构地区：[1]同济大学计算机科学与技术系,上海200092

出　　处：《计算机工程》2007年第3期208-209,212,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(60475019)

摘　　要：提出了Real-Adaboost的一种改进算法。该算法采用预先计算类Haar特征所对应弱分类器在样本空间的划分,并动态更新人脸训练样本的权值。与以往的Real-Adaboost算法比较,该算法大大缩短了训练时间,算法训练时间复杂度降到O(T*M*N),同时加速了强分类器的收敛性能,减少检测器的弱分类器数量,减少检测时间。This paper proposes a novel human face detection algorithm based on real Adaboost algorithm. Policy that calculates in advance the partitioning of Haar-like feature weak classifiers in sample input space and updating training face samples＇ weights dynamically is adopted. This algorithm reduces training time cost greatly compared with classical real-Adaboost algorithm. In addition, it speeds up strong classifier converging, reduces the number of weak classifiers and decreases detecting time.

关 键 词：人脸检测 实值Adaboost 类HAAR特征 层叠分类器 动态权值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]