3D Face recognition using Tensor Orthogonal Locality Sensitive Discriminant Analysis
Yi Jin, Qiuqi Ruan, Yizhi Wang · 2010
In this paper, a novel appearance-based method that called Tensor Orthogonal Locality Sensitive Discriminat Analysis (Tensor OLSDA) is presented for 3D face recognition. Our algorithm is motivated by the Locality Sensitive Discriminant Analysis (LSDA) algorithm, which aims at finding a projection by maximizing the margin between data points from different classes at each local area. However, LSDA is expressed in the form of 1-D vectors and this makes it difficult to estimate the intrinsic dimensionality and to reconstruct the face data. Furthermore, the object reconstruction criterion of LSDA is non-orthogonality which distorts the local geometrical structure of the data submanifold. In this paper, Tensor OLSDA as a new object reconstruction criterion is proposed to efficiently preserve the neighborhood geometrical structure according to the Locality Sensitive Discriminant Analysis (LSDA), and the locality preserving ability is enforced by computing the mutually orthogonal basis functions iteratively with tensor data representation. Experiments on CASIA 3D face database also show the impressive performance of the proposed method.