基于主动学习的平衡类鉴别分析

特定类的思想是将传统的多类特征提取和识别任务转化为多个两类问题,由此产生了类不平衡问题,影响最优鉴别特征的提取。为了解决该问题,文中提出了一种主动学习平衡类鉴别分析（ALCBD）方法。对于每个特定类,ALCBD从其对应的大类中选取它的部分近邻样本构成特定类的近邻样本集,接着将这个近邻样本集划分成与特定类相同样本数的多个子集,然后根据主动学习的思想挑选最优子集与特定类结合成为新样本集,最后用传统的线性鉴别分析（LDA）方法得到鉴别向量。基于USPS和Honda/UCSD数据库的实验表明ALCBD方法能够有效地解决类不平衡问题,并改善了识别性能。

Class-balanced Discriminant Analysis Based on Active Learning

The class-specific idea tends to recast a traditional multi-class feature extraction and recognition task into several binary class problems.In this way,the class-imbalance problem occurs,which might affect the extraction of optimal discriminant features.In order to address this problem,propose an approach named Active-Learning based Class-Balanced Discriminant analysis（ ALCBD）.For a specific class,ALCBD selects a reduced counterpart class whose data are nearest to the data of specific class,and further divides them into smaller subsets,each of which has the same size as the specific class.Then,ALCBD chooses the optimal subset according to the idea of active learning,and further combines it with the specific class to form a new sample set.Finally perform the Linear Discriminant Analysis（ LDA） on them to obtain discriminative vectors.The experimental results on the USPS and Honda/UCSD databases demonstrate that the ALCBD approach can effectively solve the class-imbalance problem,and improve the recognition performance.