多核学习融合局部和全局特征的人脸识别算法

Fusion of Local and Global Features Using Multiple Kernel Learning for Face Recognition

提出一种基于词袋模型的新的人脸识别算法.该方法将词袋模型和词袋模型的全局模式分别作为人脸图像的局部特征和全局特征描述,最后使用多核学习方法将二者进行融合.AR、FERET、CMU PIE以及LFW公开人脸数据库上的实验结果表明,本文方法能够更好的解决小样本问题,并且对人脸的表情变化、姿态变化以及面部遮挡具有更优良的鲁棒性.

A newface recognition algorithm via bag-of-words（ Bo W） is proposed. In specific,it uses Bo W and the global pattern of Bo W respectively as the local feature and global feature of face images. Multiple kernel learning is adopted to fuse the local and global features. Extensive experiments were carried out on four face databases,i. e. AR,FERET,CMU PIE and LFW. The results showthat our method can effectively solve the small training size problem and is more robust to expression changes,position variations and occlusion.