一类模糊重构度分类器

Fuzzy Classification Algorithm Based on Reconstruction Degree

作为经典的分类器,模糊K近邻分类器(FKNN),有着较为广泛的应用,且衍生出许多改进分类器。不同于模糊K近邻分类器使用简单的投票策略构建模糊集,提出一种新的分类器,即基于重构度模糊分类器。得益于对样本重构方法的改良,新的分类器避免了传统FKNN需要根据不同的数据集合调整相应的K参数的不足,而是以数据集为驱动,进而无参数调整。同时,由于重构度从结构上对噪声所产生的影响有一定抑制作用,所以该分类器对于图片噪声的鲁棒性较强。实验结果也表明基于重构度的模糊分类器在加噪声的AR人脸库上都取得了超越其他类FKNN的表现。

This paper proposes a new classifier,namely,fuzzy classifier based on representation contribution.Thanks to the improvement of the sample reconstruction method,the new classifier avoids the disadvantage that the traditional FKNN needs to adjust the corresponding K parameters according to different data sets.This method is driven by data sets and has no parameter adjustment.At the same time,because the degree of reconstruction can restrain the influence of noise from the structure,the classifier is robust to image noise.