基于协同学的人脸分类集成

提出了一种基于协同学的人脸分类集成方法.选择不同的训练样本作为原型模型,以增加原型模型的多样性;识别时,将序参量转化为后验概率,分别运用投票法和基于和的后验概率集成方法进行识别,并提出了一种改进的基于和的后验概率集成方法,来提高集成的效果.另外,将核主分量分析和协同模式识别进行结合,在运用协同模式识别之前,先采用核主分量分析获得原始图像的最优非线性表示,以提高模式的可分性,并消除图像冗余信息的影响,然后再进行协同人脸分类.对Y a le人脸库中的图像进行识别实验,结果表明所提方法的有效性,取得了比经典的协同模式识别方法和核主成分分析更好的结果.

Ensemble of multiple classifiers for face recognition based on synergetic

Ensemble of multiple classifiers for face recognition based on synergetic is proposed in this paper.Firstly different train samples are selected as prototype patterns which make the prototype patterns diversity.In the recognition stage,order parameters are converted to posteriori probability,then voting and ensemble of posteriori probability based on add rule are used respectively to get finally results;and an improved method for ensemble of posteriori probability based on add rule is also proposed in order to enhance the performance of ensemble.In addition,kernel principal component analysis and synergetic pattern recognition are combined in this paper to improve the classification results,namely,kernel principal component analysis is applied and optimal non-linear features are gotten as prototype patterns,meanwhile the effect of image redundance information is eliminated,then synergetic pattern recognition is applied for face classification.To verify the effectiveness of the proposed method,experiment is tested on Yale face database and experiment result shows that the face recognition method proposed is more available than classical synergetic pattern recognition method and kernel principal component analysis.