融合多通道信息的二维人脸识别

为充分提取人脸图像信息,进一步提高人脸识别效率,提出了一种融合图像多通道信息的二维人脸识别方法.该方法利用Haar小波变换将人脸图像变换到频率域,并获得图像4个频率域的信息;对每个频率域的图像,通过局部二值模式(LBP)进行统计编码,并提出基于HaarLBP直方图序列的人脸图像表征方法;提出2种直方图序列的匹配算法,并通过分析各个频域图像信息对识别的贡献率,进一步融合4通道图像信息进行人脸识别.在ORL和Yale人脸库上的实验结果证明,提出的识别方法对于人脸姿态、表情和光照变化有一定的鲁棒性.

Face Recognition Fusing Multichannel Information

To extract more information from face images and promote the recognition efficiency,a novel 2D face recognition approach is proposed for fusing multichannel information of face image.First,a face image is decomposed into four subimages by the Haar wavelets transform.Then,each subimage is represented by a new face representation approach,Haar local binary pattern histogram(HLBPH).Finally,two novel match models for face samples are presented to balance the different face regions.The linear weighted fusion strategy to fuse the four channels’ facial information.Experimental results are tested on ORL and Yale face database,which show that the proposed algorithm is effective and robust to the facial pose,expression,and illumination conditions.