基于改进深层网络的人脸识别算法

发布时间：2018-04-09 10:13

本文选题：人脸识别　切入点：改进的深层网络　出处：《电子学报》2017年03期

【摘要】：目前的人脸识别算法在其特征提取过程中采用手工设计(hand-crafted)特征或利用深度学习自动提取特征.本文提出一种基于改进深层网络自动提取特征的人脸识别算法,可以更准确地提取出目标的鉴别性特征.算法首先对图像进行ZCA(Zero-mean Component Analysis)白化等预处理,减小特征相关性,降低网络训练复杂度.然后,基于卷积、池化、多层稀疏自动编码器构建深层网络特征提取器.所使用的卷积核是通过单独的无监督学习获得的.此改进的深层网络通过预训练和微调,得到一个自动的深层特征提取器.最后,利用Softmax回归模型对提取的特征进行分类.本文算法在多个常用人脸库上进行了实验,表明了其在性能上比传统方法和普通深度学习方法都有所提高.
[Abstract]:In the present face recognition algorithm, handcrafted features are designed by hand or automatically extracted by depth learning in the process of feature extraction.In this paper, a face recognition algorithm based on improved deep network automatic feature extraction is proposed, which can extract the discriminative features of the target more accurately.Firstly, the image is preprocessed by ZCA(Zero-mean Component Analysis (ZCA(Zero-mean Component Analysis) whitening to reduce the feature correlation and the complexity of network training.Then, a deep network feature extractor is constructed based on convolution, pool and multi-layer sparse automatic encoder.The convolution kernel used is obtained through individual unsupervised learning.An automatic deep feature extractor is obtained by pre-training and fine tuning of the improved deep-layer network.Finally, Softmax regression model is used to classify the extracted features.Experiments on several common face databases show that the proposed algorithm has better performance than traditional methods and general depth learning methods.
【作者单位】：合肥工业大学计算机与信息学院;
【基金】：国家自然科学基金(No.61371155) 安徽科技攻关项目(No.1301b042023)
【分类号】：TP391.41

【相似文献】