加权鉴别保持投影降维的非约束人脸识别研究

非约束环境下采集的人脸图像复杂多变,因稀疏保留投影(Sparse Preserving Projection, SPP)算法没有考虑到样本的局部结构使其降维效果不理想,针对该问题,本文提出了加权判别稀疏保留投影(Weighted Discriminant Sparse Preserving Projection, WDSPP)算法。首先,引入样本类别标签和类内紧凑项,用以增强待测样本和同类样本之间的重构关系;其次,非控环境下样本质量参差不齐,考虑以样本距离权值约束稀疏重构系数,降低同类奇异样本的影响,进一步提高重构关系的准确度;最后,低维投影阶段增加全局约束因子,利用样本全局分布中隐含的鉴别信息使低维子空间分布更紧凑、更易于鉴别。在AR库、Extended Yale B库、LFW库和PubFig库上的大量实验结果表明,本文所提算法在复杂人脸环境下具有较好的识别结果。  

Unconstrained Face Recognition Based on Weighted Discriminant Sparsity Preserving Embedding

The face images acquired in the unconstrained environment are complex and changeable, and sparse preserving projection (SPP) algorithm is not ideal for dimensionality reduction. In view of this problem, a weighted discriminant sparse preserving projection for unconstrained face recognition is proposed. Firstly, in order to enhance the reconstruction relationship between query sample and training samples of the same type, and class label information of samples and the in-class compact item are added. Secondly, due to the uneven quality of the sample under non-control environment, the sample distance weight is used to constrain the sparse reconstruction coefficient, which reduces the influence of similar singular samples and further improves the accuracy of the reconstruction relationship. Finally, the global constraint factor is added in the low-dimensional projection process, and the low-dimensional subspace distribution is made more compact and more discriminative by using the implicit identification information in the global distribution of the sample. the effectiveness of the method is verified by some experiments on the AR, the Extended Yale B, the LFW and the PubFig databases.