基于多尺度LBP的人脸识别

提出一种基于多尺度LBP(Local Binary Pattern)的人脸识别算法。建立人脸图像高斯差分尺度空间,计算尺度空间图像的LBP特征,将LBP特征图像划分为互不重叠的特征区域,然后分别进行直方图统计,最后将所有区域的LBP直方图序列连接起来得到多尺度LBP特征,采用最近邻分类器对人脸图像分类识别。实验分析表明,多尺度LBP特征具有较强的人脸图像描述能力,能够提取到更加丰富的全局信息,鲁棒性强,在识别率和识别速度上均比SIFT算法高。

FACE RECOGNITION BASED ON MULTI-SCALE LBP

A face recognition algorithm based on multi-scale Local Binary Pattern(LBP) is proposed.In the algorithm,we establish the Gaussian difference scale space for face images,calculate the LBP features of the scale space images,divide the LBP feature of images into non-overlapping characteristic areas,and then conduct the histogram statistics on them respectively.Finally the multi-scale LBP features are derived by connecting the LBP histogram of sequences in all areas,and the nearest neighbour classifier is used for face image classification and recognition.Experimental analysis shows that the multi-scale LBP features of face images have strong ability of describing face images,and can extract much richer global information with strong robustness,it has a higher recognition rate and recognition ratio than those of the SIFT algorithm.