【摘要】 在人脸识别研究问题中,传统的K-近邻分类器是仅基于一种测度进行分类的。但是,这种仅基于一种测度进行分类的方法没有充分考虑不同特征间的相似信息,因而往往分类不够准确。针对这个问题,本文提出了基于距离和角度两种测度联合分类的改进近邻分类器。即在距离测度的基础上融合cosine分类器的角度信息作为分类测度,同时在分类过程中运用模糊识别,以改善传统近邻分类器的分类效果。经计算机仿真数据实验,表明这种改进的近邻分类器与Gabor小波的结合,提高了人脸识别率。更多还原

【Abstract】 The traditional K-nearest neighbor classifier is only based on one kind of measure,it doesn’ t entirely consider the sim- ilarity information between different features.This paper proposes an improved classifier based on both Euclidean distance and angle measures.It combines the angle information of the cosine classifier with the distance as the measurement criterion,and uses fuzzy pattern recognition in the classification to improve the performance of the traditional K-nearest classifier.Experimental results suggest that the combination of improved classifier proposed in this paper with Gaber wavelet achieves an improved error rate in face recognition.更多还原