Abstract: Aimed at solving the problem that 2D face recognition is sensitive to pose and illumination variations, a multi-modal face recognition approach, which is based on few feature points, is proposed. In the training stage, in order to set up the complete feature template, 3D face data are reprocessed and exploited; so as to overcome the nonlinear problem of feature extraction, a simple and effective clustering algorithm is designed, subsequently; Local Feature Analysis（LFA） is implemented to extract features of few feature points, and fusing local and global features. Experiment results confirm that the novel approach is not only efficient, but also robust to pose and illumination variations, and achieves 98.06% recognition rate on WHU-3D small-scale face database.

Key words: face recognition, multi-modal information, subset definition, few feature points

摘要： 针对二维人脸识别对姿态和光照变化较为敏感的问题，提出了一种基于少量特征点的多模态人脸识别方法。在训练阶段，对三维人脸数据进行二次处理和数据挖掘，为建立完备的特征模板奠定了基础；提出了一种简洁高效的样本聚类方法，克服了特征提取过程中的非线性问题；通过局部特征分析（Local Feature Analysis，LFA）实现了特征点“局部”与“全局”信息的融合。实验证明该方法在具有较高执行效率的同时，对人脸图像的姿态和光照变化具有理想的鲁棒性，在WHU-3D小规模人脸数据库上取得了98.06%的识别率。

关键词: 人脸识别, 多模态信息, 子集划分, 少量特征点