《学习多视角人脸概率分布模型》中学生教育文学丛书.pdfVIP

Learning Probabilistic Distribution Model for Multi-View Face Detection Lie Gu Stan Z Li Hong-Jiang Zhang Microsoft Research China, Beijing, China, 100080 Email: {liegu, szli, hjzhang}@ Abstract of the objects from possible views, which often tends to be a more complicate problem. Furthermore, the learning of Modeling subspaces of a distribution of interest in distribution is simplified to be performed only within a high dimensional spaces is a challenging problem in pat- fixed view class. In this paper, we propose a novel frame- tern analysis. In this paper, we present a novel framework work which follows the view-based scheme to learn for pose invariant face detection through multi-view face nonlinearly distributed view subspaces via the PPCA mix- distribution modeling. The approach is aimed to learn a tures model, and apply the learned representation to face set of low-dimensional subspaces from an originally detection. nonlinear distribution by using the mixtures of probabilis- tic PCA [16]. From the experiments, we found the learned Learning a subspace representation usually involves deriv- PPCA models are of low dimensionality and exhibit high ing a set of basis components from training data. As a local linearity, and consequently offer an efficient repre- powerful technique for data reduction, principal compo- sentation for visual recognition. The model is then