多模态人脸识别融合方法比较研究

Comparative research on fusion methods for multi-modal face recognition

比较研究了多模态人脸识别中的5种匹配得分级融合方法。首先用局部二值模式(Local Binary Pattern,LBP)算子分别提取人脸灰度图像和深度图像的区域LBP直方图序列(LBP Histogram Sequence,LBPHS),采用Fisherfaces分别构建相应的线性子空间,用余弦相似度计算投影向量的匹配得分,再采用5种方法对匹配得分进行融合。在FRGC数据库上的实验结果表明,除最小匹配得分外,其他融合方法的识别性能都要优于单一模态的方法。

Five fusion methods at match score level have been compared for multi-modal face recognition.Firstly Local Binary Pattern（LBP） descriptor is used to extract the LBP Histogram Sequence（LBPHS） from greyseale and depth face images.Then the corresponding linear subspaces are constructed by Fisherfaces respectively.The cosine similarity is adopted to compute the match scores of projected vectors.Then five methods are utilized to fuse match scores.The experimental results on FRGC database indicate that the recognition performance of all fusion methods except min-score is better than that of unimodal ones.