基于AFSA和PSO融合优化的AdaBoost人脸检测算法  

Ada Boost Face Detection Algorithm Based on Fusion Optimization of AFSA and PSO

作　　者：任克强[1] 高晓林[1] 谢斌[1] 

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《小型微型计算机系统》2016年第4期861-865,共5页Journal of Chinese Computer Systems

基　　金：江西省教育厅青年科学基金项目(GJJ11132)资助;江西省研究生创新基金项目(YC2013-S199)资助

摘　　要：针对传统Ada Boost人脸检测算法存在的不足,提出一种将人工鱼群算法、粒子群优化算法和Ada Boost算法相结合的人脸检测算法.该算法利用人工鱼群算法的最佳寻优特性,弥补粒子群优化算法易陷入局部最优的不足,改善粒子样本的枯竭和退化;在Ada Boost训练框架中扩展了Haar-like特征,以排除相关度较低的Haar-like人脸样本特征;采用融合优化的AdaBoost算法寻找弱分类器权重系数的最优值,组合最佳弱分类器,级联形成最终的强分类器.实验结果表明,该算法能够有效提高检测精确率、降低训练时间,取得了较好的人脸检测效果.Aiming at the shortcomings of the traditional Ada Boost face detection algorithm,this article proposes an algorithm of face detection based on AFSA,PSO and Ada Boost.The algorithm uses the optimal searching characteristics of AFSA to make up for the shortage of PSO easy to fall into local optimization,and improves particle sample depletion and degradation.In order to eliminate the lowcorrelation Haar-like face sample features,Haar-like features are extended in the Ada Boost training framework.Ada Boost algorithm of fusion optimization is used to find the optimal value of the weak classifier weighting coefficient,and combines the optimal weak classifier to form the final strong classifier.The experimental results showthat the algorithm can effectively enhance the detection accuracy rate and reduce training time,and achieves better result on face detection.

关 键 词：人脸检测 人工鱼群优化 权重系数 最优弱分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]