带相关反馈的基于深度神经网络模型的人脸检索方法  

Face Retrieval Method Base on Deep Neural Networks with Relevance Feedback

作　　者：沈旭东[1] 范守科[2] 夏海军[3] 苏金波[3] 

机构地区：[1]中国科学技术大学自动化系 [2]中国人民解放军63791部队 [3]合肥市公安局网安支队

出　　处：《计算机系统应用》2016年第1期131-135,共5页Computer Systems & Applications

基　　金：中国科学院"NGB有线无线融合应用"重点部署项目子课题(KGZD-EW-103-5(5));国家科技支撑计划子课题(2012BAH73F02);安徽省科技攻关项目(1301b042012);"核高基"重大专项(2012ZX01034-00-001)

摘　　要：针对大规模人脸检索问题,提出了一种带相关反馈的基于深度神经网络模型的人脸检索方法.首先利用卷积神经网络对人脸进行特征提取,再利用传统的检索方法进行人脸检索,在检索环节之后加入相关反馈环节.根据用户反馈的结果,将样本分成正例和负例,作为反馈环节的训练样本,完成反馈环节的训练.实验表明,该方法能够显著提高人脸检索的准确率.In this paper, a face retrieval method based on deep neural networks with relevance feedback has been presented to solve the problem of large-scale human face retrieval. Firstly, convolutional neural networks has been used for feature extracting, then, traditional search methods will be used in face retrieval. A feedback model is added after the retrieval stage. According to the users＇ feedback, result samples are divided into positives and negatives, which are be used for training the feedback model. Experiments show that this method can significantly improve the accuracy of face retrieval.

关 键 词：人脸检索 卷积神经网络 哈希检索 相关反馈 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]