鉴别相对梯度直方图特征提取与分类方法

Discriminant relative histogram feature extraction and classification

在相对梯度直方图特征的基础上,结合Fisher线性鉴别分析和角度距离相似性度量方法,提出了一种鉴别相对梯度直方图特征提取与分类方法。充分利用相对梯度直方图和鉴别分析的优势,使所得特征保留更多的对分类有利的信息;引入角度距离相似性度量,很好地克服了传统余弦相似性度量的缺陷,使人脸分类更准确。通过FERET、YaleB和PIE 3个人脸图像子集上的实验证实,鉴别相对梯度直方图特征提取与分类方法能显著提升图像梯度描述特征的分类精度,并对人脸的光照变化具有良好的健壮性。

To further improve the face recognition accuracy of image gradient features,a discriminant relative histogram feature extraction and classification method,which combines the Fisher linear discriminant analysis （FLDA）and relative gradient histo-gram features （RGHF）,is proposed in this paper.Much more classification favorable information can be preserved due to the su-periority of both RGHF and FLDA.In order to get improved face recognition performance,the angular-distance similarity meas-ure is introduced to overcome the drawbacks of traditional cosine similarity measure.Experiments on FERET,YaleB and PIE subsets indicate that the proposed method for extracting and classifying discriminant relative histogram features significantly im-proves the recognition accuracy of gradient features.Meanwhile,the proposed method is robust to the variance of illumination in face recognition.