CAJ | 학술논문

本文研究了人脸识别应用中的＂维数灾难＂问题。针对经典欧式距离不能较好刻画高维空间中样本间相似性特征,我们引入了相关系数鉴别分析方法,在高维空间中,基于相关系数提取鉴别特征,实现了人脸样本的维数约简。本文实验结果表明,该方法在小规模及大规模数据集上均具有较好的性能。
본문연구료인검식별응용중적＂유수재난＂문제。침대경전구식거리불능교호각화고유공간중양본간상사성특정,아문인입료상관계수감별분석방법,재고유공간중,기우상관계수제취감별특정,실현료인검양본적유수약간。본문실험결과표명,해방법재소규모급대규모수거집상균구유교호적성능。
This paper considers the dimensionality problem in face recognition application.The classic Euclidean distance is so simple that it can not describe the similarity relationships between samples in high-dimension space.The Correlation Discriminant Analysis method is introduced.In high-dimension space,the correlation similarity features are extracted.Finally,Random Forest is adopted as the classifier.The experiments in the paper valuate the performance of the proposed method on large databases and small databases.