基于局部Gabor变化直方图序列的人脸描述与识别

Histogram Sequence of Local Gabor Binary Pattern for Face Description and Identification

提出了一种在Gabor变换幅值域内提取局部变化模式空间直方图序列(histogramsequenceoflocalGaborbinarypatterns,简称HSLGBP)的人脸描述及其识别方法.鉴于Gabor特征对光照、表情等变化比较鲁棒,并已在人脸识别领域得到成功应用,首先对归一化的人脸图像进行多方向、多分辨率Gabor小波滤波,并提取其对应不同方向、不同尺度的多个Gabor幅值域图谱(Gabormagnitudemap,简称GMM),然后在每个GMM上采用局部二值模式(localbinarypattern,简称LBP)算子抽取局部邻域关系模式,最后由这些模式的区域直方图形成的序列来描述人脸.Gabor变换、LBP、空间区域直方图的采用使得该方法对光照变化、表情变化、误配准等具有良好的鲁棒性.而且,这种人脸建模方法不需要基于训练集合进行统计学习,因而不存在推广性问题.同时,进一步探讨了如何在分类器设计阶段与统计方法进行结合的问题,提出了统计Fisher加权的HSLGBP匹配方法.在通过FERET人脸库光照、表情和时间变化测试集上与已发表的实验结果进行对比,充分验证了该方法的有效性.

In this paper, a method for face description and recognition is proposed, which extracts the histogram sequence of local Gabor binary patterns （HSLGBP） from the magnitudes of Gabor coefficients. Since Gabor feature is robust to illumination and expression variations and has been successfully used in face recognition area. First, the proposed method decomposes the normalized face image by convolving the face image with multi-scale and multi-orientation Gabor filters to extract their corresponding Gabor magnitude maps （GMMs）. Then, the local binary patterns （LBP） operates on each GMM to extract the local neighbor pattern. Finally, the input face image is described by the histogram sequence extracted from all these region patterns. The proposed method is robust to illumination, expression and misalignment by combing the Gabor transform, LBP and spatial histogram. In addition, this face modeling method does not need the training set for statistic learning, thus it avoids the generalizability problem. Moreover, how to combine the statistic method in the stage of classification and propose statistic Fisher weight HSLGBP matching method are discussed. The results compared with the published results on FERET face database of changing illumination, expression and aging verify the validity of the proposed method.