基于动态聚类及样本筛选的人脸识别

为了综合体现训练样本的共性和个性，应用动态聚类技术，通过对于训练样本集中的同类别样本进行动态聚类，形成若干样本子集，并将这些子集的类心作为代表用于距离计算，避免了采用样本全集类心作为代表所导致的样本个性削弱，也比采用所有训练样本作为代表样本减少了存储空间和计算时间。此外，通过对于训练样本进行筛选，去除了孤立样本的影响，避免了“过拟合”现象。实验结果证明了算法的有效性。

Face recognition based on dynamic clustering and sample selection

In this paper,to integrate the commonness and individuality of the training samples,the dynamic clustering is introduced to face recognition.The training samples within a class are dynamically clustered to some subsets,and the centers of the subsets are used as representatives for the distance calculation.Thus,the sample individuality weakening due to using the center of all of the samples of each class as the representative is avoided,while the storing spending and calculation spending are reduced compared to using all of the training samples as the representatives.Also,with training sample selection,the influence of the isolated samples is removed,avoiding over-fitting.The experimental results demonstrate the efficiency of the algorithm.