摘要
极限学习机(ELM)与稀疏表示分类(SRC)算法被广泛应用于人脸识别中。ELM学习速度快,但不能很好地处理噪声图像,SRC对噪声具有鲁棒性,但计算复杂度较高。针对上述2种算法的优缺点,利用子空间追踪算法求解稀疏系数,提出一种改进的人脸识别算法,从而达到高识别率与快速的识别效果。该算法根据测试样本的ELM实际输出向量判断是否为噪声图像,干净图像直接依据ELM输出向量进行分类,噪声图像采用子空间追踪算法结合SRC框架来分类。在扩展的Yale B和ORL人脸数据库上的实验结果表明,该算法不仅识别率高,且识别速度快。
Extreme Learning Machine（ELM） and Sparse Representation based Classification（SRC） algorithm are applied to face recognition widely.ELM has speed advantage while it can not handle noise well /whereas SRC shows significant robustness to noise while it suffers high computational cost.According to the advantages and disadvantages of two algorithms,this paper proposes a hybrid approach combining extreme learning machine and Subspace Pursuit（SP） for face recognition,which incorporates their respective advantages and uses subspace pursuit method to optimize solving sparse representation coefficients in SRC.According to the analysis of ELM actual output to estimate whether the test sample is a noisy image,clean image directly uses ELM actual output to classify,and noisy image applies SP with SRC method to classify.Experimental results show that the novel algorithm has high recognition rate and speed advantage in face recognition on extended Yale B and ORL face database respectively.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第1期168-173,共6页
Computer Engineering
基金
国家自然科学青年基金资助项目(61202439)
湖南省教育厅优秀青年基金资助项目(12B003)
湖南省交通运输厅科技计划基金资助项目(201334)
关键词
人脸识别
极限学习机
稀疏表示
稀疏编码
子空间追踪
face recognition
Extreme Learning Machine（ELM）
sparse representation
sparse coding
Subspace Pursuit（SP）
