基于PSO的改进AdaBoost人脸检测算法  

Improved AdaBoost face detection algorithm based on particle swarm optimization

作　　者：张均 叶庆卫[1] ZHANG Jun;YE Qingwei(College of Information Science and Engineering,Ningbo University,Ningbo Zhejiang 315211,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《计算机应用》2020年第S01期61-64,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(51675286,61071198)。

摘　　要：针对传统的AdaBoost分类算法弱分类器性能差、训练时间长的问题,提出一种基于粒子群寻优(PSO)的AdaBoost分类算法。首先,采用双阈值的弱分类器代替原始的单阈值弱分类器,建立新的弱分类器结构;其次,通过粒子群寻优的方式搜索最优特征和两个最优阈值代替传统的枚举搜索方式;最后,将所有弱分类器组合成强分类器。通过理论分析和仿真实验表明,与传统枚举方式建立的双阈值强分类器相比,基于粒子群寻优算法建立强分类器的方法,其平均训练时间缩短至原来的1/57,而且两种方式训练的强分类器性能相当。实验结果表明,基于粒子群寻优的AdaBoost算法能够有效提高分类器训练效率。For the poor performance and the long training time of the traditional AdaBoost classification algorithms,a new AdaBoost classification algorithm was proposed based on Particle Swarm Optimization(PSO).Firstly,a new weak classifier structure was established by replacing the original single threshold weak classifier with a double threshold weak classifier.Secondly,searching the optimal feature and two optimal thresholds by PSO were used ro replaece the traditional enumeration search method.Finally,all weak classifiers were combined into a strong classifiers.The theoretical analysis and simulation results show that the average training time of the method based on PSO is reduced to 1/57 of the traditional enumeration method,while the performance of the two methods is equivalent.The experimental results show that the AdaBoost algorithm based on PSO can improve the training efficiency of classifier.

关 键 词：人脸检测 ADABOOST算法 双阈值 粒子群寻优 弱分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]