色度马氏距离图与灰度图特征自适应融合的彩色人脸识别  

Color face recognition using adaptive feature fusion based on chroma mahalanobis distance map and gray map

作　　者：崔法毅[1] 

机构地区：[1]燕山大学河北省测试计量技术及仪器重点实验室,河北秦皇岛066004

出　　处：《红外与激光工程》2015年第4期1382-1389,共8页Infrared and Laser Engineering

基　　金：秦皇岛市科学技术研究与发展计划(2012021A004)

摘　　要：色彩提供了比灰度更为丰富的信息,鉴于彩色人脸图像所包含的鉴别信息远多于灰度人脸图像,将色度马氏距离图引入彩色人脸识别中。基于YCb Cr颜色空间,分离彩色人脸图像的色度与亮度信息,构建出基于色度信息的马氏距离图,同时分离出基于亮度信息的灰度图。提出一种色度马氏距离图与灰度图特征自适应融合的人脸识别算法。分别构造出色度马氏距离图与灰度图的基于小波包结点能量的归一化特征向量,采用多种融合策略进行特征融合,并根据融合效果自适应地选取特征融合参数,构造出最佳的鉴别特征向量,实现色度与亮度特征的互补。使用基于方差相似度的分类器获得人脸识别结果。实验表明:该算法识别率高、鲁棒性好。The color provides much more information than the gray. Considering that the color facial image contains much more identification information than the gray facial image, the chroma Mahalanobis distance map was introduced into color face recognition. Based on the YCb Cr color space, the information of chroma and brightness for one color facial image can be separated. Then Mahalanobis distance maps based on chroma information were constructed, at the same time, gray maps based on brightness information can be extracted by orignal color facial images. An algorithm for color face recognition using adaptive feature fusion based on chroma Mahalanobis distance map and gray map was presented. Based on energy values of wavelet packet sub-nodes, normalized feature vectors of chroma Mahalanobis distance maps and gray maps were constructed, respectively. Then identification feature vectors were constructed using several feature fusion methods, and fusion parameters were selected adaptively according to fusion effects. So feature complementation of chroma and brightness was achieved. Results of face recognition were obtained by the classifier based on the variance similarity degree. Experiments show that the algorithm has the characteristics of high recognition rate and good robustness.

关 键 词：彩色人脸识别 色度马氏距离图 YCBCR颜色空间 小波(包)变换 自适应特征融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]