Abstract: In order to efficiently integrate multi-view information and make optimal view is dominant in multi-view intact subspace learning, a multi-view collaboration intact subspace learning scheme is proposed. Furthermore, aiming to enable the intact feature representations of objects having more powerful discriminability in latent intact subspace, Fisher discriminant analysis is introduced into intact subspace learning. By employing Fisher discriminant analysis, the within-class scatter of intact feature representations is minimized and the between-class scatter of intact feature representations is maximized, simultaneously. By combining multi-view collaboration intact subspace learning scheme and Fisher discriminant analysis together, an algorithm named robust multi-view collaboration intact discriminant subspace learning is proposed. Experiment results demonstrate that the proposed algorithm can efficiently integrate multi-view information and mine discriminant information. And the proposed algorithm is an effective multi-view intact discriminant subspace learning algorithm.

Key words: multi-view learning, linear discriminant analysis, face recognition, subspace learning

摘要： 为了有效地融合多视图信息并使有利于多视图完整子空间学习的视图主导多视图学习，提出了多视图协同完整子空间学习策略。进一步，为了使对象在潜在完整子空间中的完整特征表示具有更好的鉴别能力，将Fisher鉴别分析引入到了多视图完整子空间学习中。Fisher鉴别分析可以在最小化对象的完整特征表示的类内散度的同时最大化对象的完整特征表示的类间散度。将多视图协同完整空间学习策略和Fisher鉴别分析融合在一起，提出了鲁棒多视图协同完整鉴别子空间学习算法。实验结果表明，所提算法能够有效地融合多视图信息并挖掘鉴别信息，是一种有效的多视图完整子空间学习算法。

关键词: 多视图学习, 线性鉴别分析, 人脸识别, 子空间学习