半监督鉴别特征抽取及人脸识别应用研究

项目名称： 半监督鉴别特征抽取及人脸识别应用研究

项目编号： No.60875004

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 陈才扣

作者单位： 扬州大学

项目金额： 28万元

中文摘要： 当面对高维小样本情况，现有的基于少量已知类别数据样本的监督特征抽取方法和基于未知类别数据的非监督特征抽取方法会导致过拟合问题。近年来，半监督学习受到人们的普遍关注，成为机器学习和模式识别等领域的研究热点。尽管相当数量的研究报道表明，该方法优于传统的监督和非监督学习方法。但有关将半监督学习思想应用于鉴别特征抽取的研究报道较少。因此，如何将半监督学习应用于现有的监督和非监督特征抽取算法、揭示三者之间的内在联系、发掘半监督特征抽取方法的潜力，提高它在人脸识别应用中的鉴别能力，是一个有待深入研究的问题。该研究对于促进半监督学习方法本身的发展，及其在人脸识别方面的更为成功的应用，都具有重要的理论和实际意义。该项目研究的内容包括：研究建立半监督学习在特征抽取中的理论框架；理清监督、非监督和半监督特征抽取三者之间的联系；提出一系列有效的和快捷的面向分类的半监督特征抽取算法。

中文关键词： 特征抽取；半监督学习；人脸识别；模式分类

英文摘要： For high-dimensional and small size sample problems, the existing supervised feature extraction methods with a small amount of labelled data samples and unsupervised feature extraction methods with unlabelled data samples can often lead to an over-fitting problem. Semi-supervised learning has attracted a significant amount of attention in recent years and has become a hot topic in the fields of machine learning and pattern recognition. Although a great number of research reports show that the semi-supervised learning is superior to the traditional supervised and unsupervised learning methods, there are very few researchs focusing on the application of the semi-supervised learning tothe discriminant feature extraction. Therefore, it is a problem worth being explored and in-depth studied in future that how to apply semi-supervised learning to the current supervised and unsupervised feature extraction algorithms, reveal the inherent relation among them, tap the potential of the semi-supervised feature extraction methods and improve the semi-supervised discriminative ability in the application of face recognition. This research has an important theory and realistic meanings in the promotion of the development of the semi-supervised learning method and the more successful application of face recognition. The contents of this research project include (1) construction of the thoery framework of the application of the semi-supervised learning to feature extraction, (2) revealing the inherent relation between supervised, semi-supervised and unsupervised feature extraction methods, (3) developing a series of effective and efficient and classification-oriented semi-supervised feature extraction methods.

英文关键词： supervised feature extraction; unsupervised feature extraction; semi-supervised learning; face recognition