集成多分类器的人脸识别

Face Recognition by Combining Multiple Classifiers

提出一种用组合多分类器融合局部信息进行人脸识别的方法。人脸识别过程中图像样本间的相似度可建模为“类内差”和“类间差”两种模式类,用这种思想在图像小波分解域的局部区域上构造弱分类器集,然后通过Boosting训练生成强分类器,最终的人脸匹配由多个弱分类器输出的加权和给出决策。实验结果表明,系统具有较高的识别率,对表情和光照变化具有很好的鲁棒性,而且对新个体有较好的扩展能力。

This paper proposes a face recognition method by combining multiple classifiers for information fusion. The similarity between pairs of faces can be modeled as two classes,intra-pattern and inter-pattern. Firstly this idea is used to construct weak learners in local area of wavelet domains. Then the boosting algorithm is used to train the strong classifiers. The final decision of matching is given by weighted combination of multiple weak classifiers. The experimental results show that the system is robust for variation of expression and illumination. The pretty high recognition rate can be achieved even on the new data set in which the individuals are unseen during training process.