关键词: 表情流形, 保局投影, 高阶奇异值分解, 人脸识别, 表情识别

Abstract: In this paper，expression decomposition algorithm based on appearance manifolds is proposed.Firstly，face images are mapped into low-dimensional manifold subspace using Locality Preserving Projections（LPP）.Secondly，these low-dimensional representations are decomposed into person subspace and facial expression subspace.Finally，face and facial expression recognition are realized.The problem of image feature extraction in higher-order singular value decomposition is solved by using appearance manifolds learning method.By applying Higher-Order Singular Value Decomposition（HOSVD） to appearance manifolds，person pattern hardly influence facial expression recognition.The advantages of higher-order singular value decomposition and manifolds learning are fused effectively.The algorithm is tested on JAFFE and CMP-AMP face database.The experimental result shows that it is powerful and effective for face and facial recognition.

Key words: appearance manifolds, Locality Preserving Projections（LPP）, Higher-Order Singular Value Decomposition（HOSVD）, face recognition, expression recognition