基于局部彩色二元模式纹理特征的人脸分类方法

A Human Face Classification Approach on Local Color Binary Patterns

针对现有的局部纹理特征在光照变化下的人脸分类准确率不高的问题,提出一种局部彩色二元模式纹理特征提取和表示方法.该方法首先求取多个信号通道的彩色向量的模,以减少单独对每个信号通道提取LBP特征导致的量化误差;其次提取任意两个信号通道之间的像素彩色夹角来减少光照变化的影响.对于给定的测试人脸图像集,通过SVM的二分类器投票得到人脸图像的分类.在Color FERET和XM2VTSDB人脸图像数据库上的实验证明该方法在光照变化下可以有效地分类人脸图像.

Aiming at low classification rate of local texture features for illumination human face classification, this paper proposed a local color binary patterns method for texture feature extraction and presentation. We first compute the norm of color vector between multi -- channels to reduce quantify errors caused by extracting LBP features from each channel individually. Then, we extract the pixel color angular between each two channels to reduce illumination influence. Given test human face image dataset, the face is classified by SVM. The experimental results on standard dataset Color FERET and XM2VTSDB demonstrate the effectiveness of our method.