基于自适应判别局部块对齐的人脸识别

FACE RECOGNITION BASED ON ADAPTIVE DISCRIMINATIVE LOCAL PATCH ALIGNMENT

针对有监督学习下的人脸识别问题,提出自适应判别局部块对齐SALDA(Self adaptive Local Discriminative Alignment)算法用于提取人脸特征。SALDA算法利用各样本点所具有的独特的局部近邻点分布,通过同类近邻点自动构造各样本点的局部邻域;基于已构造的局部邻域,SALDA提出一个自适应局部判别分析模型,所得到的局部判别信息通过全局排列转化为统一的全局特征表示。SALDA算法具有自适应构造局部邻域和自适应局部判别分析两个特点。通过在人脸数据库上的仿真实验,证明了所提出的SALDA算法在人脸识别上的有效性。

A novel algorithm,named self-adaptive local discriminative alignment（SALDA） is presented in this paper for extracting face features in light of face recognition problem in supervised learning. SALDA algorithm makes use of unique local neighbour point distribution of each sample point to construct through similar neighbour points the local neighbourhood of each sample point; Based on the constructed local neighbourhood,SALDA builds up an adaptive local discriminant analysis model,the local discriminant information obtained will convert to the uniform global feature representation through global alignment. SALDA algorithm has two characteristics of adaptive local neighbourhood construction and adaptive local discriminant analysis. Through simulation experiment on face database we prove the effectiveness of SALDA algorithm proposed in face recognition.