Levy变异ABC算法优化二阶Volterra核的鲁棒人脸识别  

Robust face recognition based on the second order Volterra kernal optimized by ABC algorithm with Levy mutations

作　　者：王建玺[1] 王刘涛[1] 李小红[2] 

机构地区：[1]平顶山学院,河南平顶山467000 [2]武汉大学,武汉430072

出　　处：《计算机应用研究》2015年第2期619-622,626,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(40971219);河南省科技厅科技发展计划项目(134300510037)

摘　　要：针对人脸识别中由于人脸表情、姿态、尺度、光照和其他环境参数变化而影响识别性能的问题,提出了一种随机优化算法。首先,将原始图像划分成特定空间子块,并使用二阶Volterra核寻找非线性函数映射;然后,使用人工蜂群算法获取最优Volterra核,从而在特征空间内最大化类间距离并最小化类内距离;最后,利用投票策略和最近邻分类器完成人脸的分类。在两个通用人脸数据集Yale A和扩展Yale B上对该算法进行了评估,并将其与其他统计学习算法和几种最新提出的方法进行了比较。实验结果表明了Levy变异人工蜂群算法优化Volterra核的有效性,识别效果明显优于许多现有算法。The recognition performance of face recognition was affected by the facial expression, pose, scale, illumination and other environmental parameters, for which this paper proposed a novel stochastic optimization method. Firstly, it divided the original image into specific space within the block, and used the two order Volterra core for the nonlinear mapping function. Then, it used artificial bee colony optimization technique to obtain the optimal Volterra nucleus. Volterra nucleus could be optimal in the feature space to maximize the between class distance and minimize the within class distance. Finally,it used voting strategy and nearest neighbor classifier to classify the blocks. It evaluated the application of the algorithm in the two common benchmark face recognition data sets, and compared face recognition in other statistical learning algorithm and the newly proposed several methods＇ performance. Experimental results show that the artificial bee colony optimization technique with Levy mutation is in optimizing the effectiveness of Volterra nucleus, and is significantly better than many existing algorithms.

关 键 词：二阶Volterra核 鲁棒人脸识别 Levy变异 人工蜂群算法 最近邻分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]