基于分布估计算法的人脸Haar特征选择  

Haar feature selection of face detection based on estimation of distribution algorithm

作　　者：陈中胜[1] 李春贵[2] 蒋凯[1] 

机构地区：[1]广西科技大学电气与信息工程学院,广西柳州54500 [2]广西科技大学计算机学院,广西柳州54500

出　　处：《广西科技大学学报》2015年第2期47-52,共6页Journal of Guangxi University of Science and Technology

基　　金：广西自然科学基金(2011gxnsfa018162)资助

摘　　要：通过研究人脸检测算法中Ada Boost算法,针对算法中的haar特征维数过高、训练耗时过长,检测效率过低等问题.提出基于分布估计算法(Estimation of Distribution Algorithm,EDA)的人脸haar特征选择人脸检测.EDA采用类内类间比作为适应度函数,通过统计学习的手段建立解空间内个体分布的概率模型,然后对概率模型随机采样产生新的群体,进行反复计算,实现群体的进化,最终得到全局最优解,以此来实现haar特征选择.实验结果表明:检测率(DR)与误检率(FDR)优于传统算法,而且检测速度得到了提升.Through the study of face detection based of AdaBoost algorithm,in view of the problem of the algorithm with high dimension,redundancy and being time consuming of Haar features,which seriously impact on the speed of face detection.A facial feature selection method based on estimation of distribution algorithm （EDA） was proposed. EDA uses the ratio of intra-class and inter-class as the fitness function, by means of statistical learning probabilistic model to establish the distribution of the solution space of individual distribution, then random sampling the probability model to generate new groups, so repeatedly, to achieve the evolution of populations, and ultimately gets the global optimal solution, finally achieves Haar feature selection. Experimental results show that the detection rate （DR） and false detection rate （FDR） is superior to the traditional algorithm, and the detection rate has been improved.

关 键 词：人脸检测 HAAR特征 分布估计 特征选择 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]