一种改进的Adaboost人脸检测方法  

An improved Adaboost method for face detection

作　　者：张志勋[1] 张磊[1] 杨凡[1] 

机构地区：[1]兰州工业学院软件工程系

出　　处：《自动化与仪器仪表》2013年第6期143-145,148,共4页Automation & Instrumentation

基　　金：甘肃省高等学校研究生导师科研项目计划资助(0914-03);甘肃省科技支撑项目计划资助(1204GKCA043)

摘　　要：针对现有基于粒子群(PSO)策略的Adaboost人脸检测方法没有考虑到PSO容易陷入局部最优且后期收敛速度较慢的问题,提出一种改进的Adaboost人脸检测方法。该方法将自适应逃逸粒子群(AEPSO)引入传统Adaboost人脸检测中,利用粒子表达Haar-Like矩形特征,从而将特征选择和分类器构建转化为AEPSO问题进行解决。基于Matlab仿真实验的结果表明,改进后的方法具有较好的检测性能。Existing Adaboost methods for face detection based on particle swarm optimization （PSO） does not consider that PSO suffers from defects of trapping in local optimum and slow convergence, to solve this problem, an improved Adaboost method for face detection is proposed. In this method, self-adaptive escape PSO （AEPSO） is introduced into conventional Adaboost face de- tection, meanwhile, Haar-Like rectangular features are expressed as particles, so that features selection and classifiers construction could be resolved by using AEPSO. Results of simulation based on Matlab indicated the improved method obtains better detection performance.

关 键 词：人脸检测 PSO ADABOOST AEPSO Haar-Like矩形特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]