Gabor特征与深度信念网络结合的人脸识别方法

提出了一种基于Gabor特征和深度信念网络(DBN)的人脸识别方法,通过提取Gabor人脸图像的不同尺度图进行卷积融合,将融合后的特征图作为DBN的输入数据,训练多层来获得更加抽象的特征表达,整个训练的过程中采用交差熵来微调DBN,模型的最顶层结合Softmax回归分类器对抽取后的特征进行分类.在AR人脸库测试的实验结果表明:将Gabor特征与DBN结合应用于人脸识别,其准确率可高达92.7％,与其他浅层学习模型相比,DBN学习了数据的高层特征的同时还降低了特征维数,提高了分类器的分类精度,最终有效改善了人脸识别率.

Face recognition algorithm based on Gabor feature and DBN

A method for face recognition based on the Gabor feature and deep belief network (DBN)is proposed.By extracting different scales image of Gabor face images for convolution fusion and fused feature image is used as input data of DBN.Many layers are trained in order to get more abstract representation.In whole training process,cross entropy method is adopted to fine-tune DBN.The Softmax regression classifier is used for classification which is implemented at the top layer.The experimental result in AR face database shows that when Gabor feature extract combining with DBN are applied to face recognition,its accuracy reaches 92.7 ％.Comparing with other shallow-layer learning models,DBN not only studies the high-level features of the data,but also reduces dimension and improves the precision of the classifier,which finally improves face recognition rate.