摘要： 研究快速多分类器集成算法。对多分类器集成需选定一定数量的弱分类器，再为每个弱分类器分配一定权重。在选择弱分类器时，通过计算每个弱分类器在全部训练样本集上的分类错误率，对其进行排序，挑选出分类效果最好的若干弱分类器。在多分类器权重分配策略上，提出2种权重分配方法：Biased AdaBoost算法与基于差分演化的多分类器集成算法。在人脸数据库上的实验结果表明，与经典AdaBoost算法相比，该算法能有效降低训练时间，提高识别准确率。

关键词: 快速多分类器集成, 差分演化, AdaBoost算法, 人脸识别, 训练时间

Abstract: This paper presents the research on fast Multi-classifier Ensemble(MCE) algorithm. MCE gets a number of classifiers, and assigns weights to the classifiers. A certain number of best classifiers can be gotten based on the error rate of every classifier. Assigning the weight of classifier is researched on, and two training methods are presented. The first is Biased AdaBoost algorithm which is sequentially to compute the weight of classifier. The second is DE-MCE based on Differential Evolution(DE) algorithm which optimizes the weights of all selected classifiers. Experimental result on face recognition shows that the training time of the algorithm is better than AdaBoost algorithm, and has high accuracy rate.

Key words: fast Multi-classifier Ensemble(MCE), Differential Evolution(DE), AdaBoost algorithm, face recognition, training time

中图分类号: 

• TP18