基于多尺度训练库与多特征融合的人脸识别

针对光照差异、表情变化、遮挡等因素造成人脸识别率低的问题,提出一种基于多尺度训练库和加权特征的鲁棒性人脸识别算法。首先根据不同大小的图片具有不同信息量的特点定义并建立多尺度训练库,然后采用RPCA方法对人脸图像进行分解,之后进行HMLBP特征和Eigenface特征提取,最后引入一个权重因子将两种特征进行加权融合,并采用基于稀疏表达的方法对人脸图像进行识别。实验结果表明,相比其他人脸识别算法,本文提出的算法对标准人脸库保持较高识别率,最高可达99%,同时对遮挡人脸库也具有较好的识别效果,鲁棒性较高。

Face Recognition by features fusion based onMultiscale Training Set

For the low face recognition rate caused by frontal views with varying expression, illumination and occlusion, an face recognition algorithnm with good robustness based on multiscale traning set and weighted features is proposed. Firstly, the algorithnm establishes a multiscale training set according to different size of images which contains different information. Next, images are decomposed by using RPCA. Finally, the HMLBP features and the eigenface features are weighted combined for face recognition based on sparse representation. Experiments show that, compared with other algorithms, the proposed algorithnm has a high recognition rate which can be 99% and has high robustness whatever it is based on common face database or occluded faces.