An incremental learning algorithm of Recursive Fisher Linear Discriminant
An incremental learning algorithm of Recursive Fisher Linear Discriminant
复制标题
DOI：
10.1109/ijcnn.2009.5178963
复制
发表时间：
2009-06
期刊：
影响因子：
--
通讯作者：
Ryohei Ohta;S. Ozawa
中科院分区：
文献类型：
--
作者：
Ryohei Ohta;S. Ozawa
本文提出了一种在线特征提取方法-增量递归Fisher线性判别分析（IRFLD），其批量学习算法RFLD是由Xiang等人提出的。在传统的线性判别分析（LDA）中，由于类间散布矩阵的秩，判别向量的数目被限制为类数减1。RFLD和建议的IRFLD可以消除这种限制。在拟议的IRFLD，庞等人。的增量线性判别分析（ILDA）进行了扩展，使得有效的判别向量递归搜索的互补空间的常规ILDA子空间。此外，为了估计合适数量的有效鉴别向量，我们还提出了一个递归计算的收敛准则，该准则通过使用投影在互补子空间上的鉴别特征的类可分性来定义。实验结果表明，随着学习的进行，IRFLD的识别准确率不断提高。对于几个数据集，我们证实，建议IRFLD优于ILDA的识别精度。然而，IRFLD相对于ILDA的优势取决于数据集。
This paper presents an online feature extraction method called Incremental Recursive Fisher Linear Discriminant (IRFLD) whose batch learning algorithm called RFLD has been proposed by Xiang et al. In the conventional Linear Discriminant Analysis (LDA), the number of discriminant vectors is limited to the number of classes minus one due to the rank of the between-class scatter matrix. RFLD and the proposed IRFLD can eliminate this limitation. In the proposed IRFLD, the Pang et al.'s Incremental Linear Discriminant Analysis (ILDA) is extended such that effective discriminant vectors are recursively searched for the complementary space of a conventional ILDA subspace. In addition, to estimate a suitable number of effective discriminant vectors, we also propose a convergence criterion for the recursive computations which is defined by using the class separability of discriminant features projected on the complementary subspace. The experimental results suggest that the recognition accuracies of IRFLD is improved as the learning proceeds. For several datasets, we confirm that the proposed IRFLD outperforms ILDA in terms of the recognition accuracy. However, the advantage of IRFLD against ILDA depends on datasets.