摘要： 提出一种在分数层上对全局和局部特征进行融合的人脸识别方法。全局特征由不同局部描述算子对整幅人脸图像进行运算产生，局部特征按空间位置的不同划分由直接抽取全局特征的子集构成。根据实际应用中对人脸识别系统速度和精度的不同要求，给出2种融合策略组合全局和局部特征。在FRGC v2.0大规模人脸库上的实验结果表明，该方法在增加少量运算的条件下能使系统性能明显提升。

关键词: 人脸识别, 全局特征, 局部特征, 融合, 分数层

Abstract: This paper proposes a score level fusion method using both global and local features for face recognition. The global features are constructed by applying several local descriptors to the whole face image. The local features are obtained directly from the subsets of the constructed global features with corresponding space locations. Two fusion strategies are proposed based on different practical preferences on the performance or computational time of the face recognition system. Experimental results on the large scale face database(FRVT v2.0) show that the proposed methods significantly improve system performance with small computational time increments.

Key words: face recognition, global feature, local feature, fusion, score level

中图分类号: 

• TP18