• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 图形与图像 • 上一篇    下一篇

基于改进灰狼算法优化支持向量机的人脸识别

冯璋,裴东,王维   

  1. (西北师范大学物理与电子工程学院,甘肃 兰州 730070)
  • 收稿日期:2018-07-03 修回日期:2018-10-17 出版日期:2019-06-25 发布日期:2019-06-25

Face recognition by support vector machine
optimized by an improved grey wolf algorithm

FENG Zhang,PEI Dong,WANG Wei   

  1. (School of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730070,China)
  • Received:2018-07-03 Revised:2018-10-17 Online:2019-06-25 Published:2019-06-25

摘要:

针对二维主成分分析法(2DPCA)与主成分分析法(PCA)相结合提取人脸特征时效率不高的问题,提出一种2DPCA和快速PCA结合与改进灰狼算法(EGWO)共同优化支持向量机的人脸识别方法。该方法在特征提取方面运用2DPCA与快速PCA相结合,以减少提取特征的维数和提取时间,从而缩短了SVM所需的识别时间。为了提高灰狼算法的全局搜索能力,引用精英反向学习策略初始化种群个体,有效增强GWO的勘探和开采能力,再将其使用到SVM中,迭代获取最佳核参数和惩戒参数,将训练得到的最终分类器应用于人脸识别中。通过6个基准测试函数与GWO和反向学习灰狼算法(OGWO)进行性能比较,改进灰狼算法的收敛精度和收敛速度更优;经ORL和Yale中的人脸图像实验,证明了改进算法相对于GWO、粒子群算法(PSO)和差分进化算法(DE)结合SVM模型的识别结果更佳且稳定性更强。

 

关键词: 人脸识别, 主成分分析, 灰狼算法, 支持向量机, 精英反向学习

Abstract:

In order to solve the low efficiency problem of the combined twodimensional principal component analysis (2DPCA) and principal component analysis (PCA) in extracting face features, we propose a face recognition method based on the support vector machine optimized by the 2DPCA and fast PCA combined with an improved gray wolf algorithm (EGWO). Concerning the feature extraction, the method combines the 2DPCA with fast PCA to reduce the dimension and extraction time of the extracted features, thus reducing the identification time required by the SVM. In order to improve the global search capability of the gray wolf algorithm, the elite opposite-based learning strategy is used to initialize population individuals, which effectively enhances GWO’s exploration and mining capabilities. And then the elite opposite-based learning strategy is used in the SVM to iteratively obtain the best kernel parameters and disciplinary parameters, and the final classifier obtained from training is applied in face recognition. The GWO and opposite-based learning grey wolf algorithm (OGWO) are compared on six benchmark test functions in convergence accuracy and speed, and the former outperforms the latter. Experiments on face images in ORL and Yale datasets show that the improved algorithm is better and more stable than the GWO, particle swarm optimization (PSO) and differential evolution (DE) algorithm combined with SVM model.
 
 

Key words: face recognition, principal component analysis (PCA), grey wolf algorithm, support vector machine, elite opposite-based learning