Manifold Recognition Based on Discriminative-Analysis of Canonical Correlations
Hui Jie Zhang, Chen Wang, Hong Guang Sun, Yan Wen Li, Wei Guan · Advanced materials research · 2011
Nowadays, the idea of recognition based on image sets looms so large in real world applications. From the view of manifold learning, each image set has commonly been regarded as a manifold, and we formulate the problem of set recognition as manifold recognition (MR). Since it is impossible to directly compute the distance between nonlinear manifolds, constructing local linear subspaces is brought into our focus. Among methods offering of subspace matching, canonical correlations have recently drawn intensive attention. For the task of MR, we propose a method of Manifold Recognition Based On Discriminative-analysis of Canonical Correlations (MRDCC). The proposed method is evaluated on two datasets: Honda/UCSD face video database and ETH-80 object database. Comprehensive comparisons and results demonstrate the effectiveness of our method.