人脸检测级联分类器快速训练算法  

An Improved Algorithm for Constructing Cascade Classifier Based on AdaBoost

作　　者：唐徙文[1] 曾义[1] 

机构地区：[1]上海交通大学计算机科学与工程系,上海200240

出　　处：《计算机仿真》2007年第12期324-327,共4页Computer Simulation

摘　　要：目前AdaBoost训练算法已被广泛地应用于人脸检测中级联分类器的构建,而AdaBoost算法训练级联分类器的周期却十分漫长。为了减少训练时间,文中提出了一种基于AdaBoost的改进训练算法。该算法通过对弱分类器的阈值选择进行一趟处理来降低运算时间复杂度,并根据AdaBoost训练迭代中只改变样本权值而不更新样本的特点对特征值和排序结果进行缓存来提高训练算法的性能。实验结果表明,该算法大幅提高了人脸检测分类器训练系统的性能,使得分类器的训练时间缩短了60多倍。由于AdaBoost算法的通用性,该改进算法不仅适用于人脸检测,也适合所有进行权值更新迭代训练的Boosting算法。AdaBoost training algorithm has been broadly used in constructing cascade classifier for face detection.However,the training period is unbearably long.An improved training algorithm of AdaBoost is proposed to solve this problem in this paper.According to this improved algorithm,threshold selection of weak classifiers is processed in a single pass to reduce time complexity.And computational results of sorted feature values are buffered,cutting down the number of re-calculations significantly.Experimental results indicate that these improvements promoted more than 60 times performance of training system.Apart from face detection,this algorithm can be applied in various environment involving iterative weights adjusting training.

关 键 词：适应性自益算法 级联分类器 高性能训练算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]