摘要
人脸识别是统计模式识别领域中经典的分类问题,为了提高算法的分类性能,优化技术被广泛应用到人脸识别领域。提出基于稀疏恢复的l1范数凸包分类算法,将原始训练数据集进行低秩恢复,利用恢复出的低秩矩阵和误差矩阵构成新训练集字典建立各类训练样本凸包模型,并在l1范数意义下,计算观测样本与各类凸包模型差值,用所得差值等价观测样本到各类样本凸包的距离,将距离最小的一类视为判别输出类。在ORL(Olivetti Research Lab)标准人脸图像库上进行实验分析,实验证明基于稀疏恢复的l1范数凸包分类算法具有较高的识别效率。
In order to improve the classification performance of the algorithms,optimization technology is widely used in the field of face recognition,which is considered as a classical classification problem of statistical pattern recognition. In this paper,an l1 norm convex hull classification algorithm based on sparse recovery is put forward. Lowrank matrix is recovered from the original training data,and then convex hulls of different training-sample models are established in the light of the newtraining set dictionary composed of the recovered lowrank matrix and error matrix. In the sense of l1 norm,the difference between test sample and various kinds of convex hull model is calculated. The results are then used to test the distance between sample and various kinds of convex hull,and the category with smallest distance is considered as discriminant output class. The experimental analysis conducted on the ORL( Olivetti Research Lab) face image database shows that l1 norm convex hull classification algorithm based on sparse recovery has higher recognition efficiency.
出处
《沈阳航空航天大学学报》
2016年第1期42-46,共5页
Journal of Shenyang Aerospace University
基金
国家自然科学基金面上项目(项目编号:11371255)
辽宁省高等学校优秀科技人才支持计划(项目编号:LR2015047)
关键词
人脸识别
L1范数
凸包
距离
face recognition
l1 norm
convex hull
distance