The goal of one-class classification model is to design the covering model for the target class. The model should ideally cover the objects of target class, and reject all other non-targets. The support vector data description (SVDD) aims to find the minimum ball enclosed all the target objects. However, this model couldn't perform better for the data with irregular and complex distribution. The convex hull data description (CHDD) based tight covering model is presented in this paper. The model is a non-parametric classifier which covers the irregular data adaptively in the feature spaces. By the introduction of kernel functions, the stronger ability of nonlinear classification could be obtained. When the training set of target class contains outliers, the model can be made more robust by rejecting a fraction of the training objects. The experimental results show that the presented method performs better by comparing its results on the UCI, MNIST and MIT-CBCL face data sets with other one-class cl assification methods.%一类分类问题的研究目标是设计目标类样本的覆盖函数,理想情况下使得目标类样本被接受,所有非目标类的样本被拒绝.经典SVDD覆盖模型寻找包含训练数据的最小半径超球对其进行覆盖,该模型对非规则复杂分布的数据描述存在较多的冗余区域.本文提出一种基于训练集样本凸壳数据描述(Convex Hull Data Description,CHDD)的紧致覆盖模型.该模型无须参数设置,可实现对样本非规则复杂分布的自适应覆盖,并可通过利用核函数方法获得更强的非线性分类能力.当训练集包含噪声样本时,通过拒绝一定比例的目标类样本可获得更为鲁棒的凸壳边界描述.在UCI数据库、MNIST手写体数据库和MIT-CBCL人脸识别数据库上的实验结果表明了本文方法的有效性,相比现有一类分类算法,CHDD取得更好的分类效果.
展开▼
