人工智能培训

搜索

机器学习培训:面向识别的成对关系网络(Pairwise Relational Networks for Face Recognition)

[复制链接]
mhtq 发表于 2018-8-16 09:42:02 | 显示全部楼层 |阅读模式
mhtq 2018-8-16 09:42:02 2934 0 显示全部楼层
机器学习培训:面向识别的成对关系网络(Pairwise Relational Networks for Face Recognition)使用深度神经网络的现有面部识别很难知道用什么类型的特征来清楚地区分面部图像的身份。为了研究人脸识别的有效特征,我们提出了一种新的人脸识别方法,称为成对关系网络(PRN),它获取特征图上标志点周围的局部外观补丁,并捕获一对局部外观补丁之间的成对关系。 PRN经过训练以捕获不同身份之间的独特和判别性成对关系。由于空间关系的存在和意义应该是依赖于身份的,我们添加一个人脸身份状态特征,它从长期短期记忆(LSTM)单位网络中获得特征图上的连续局部外观补丁到PRN。进一步改善面部识别的准确性,我们结合了globalappearance表示与成对关系特征。 LFW的实验结果表明,仅使用成对关系的PRN达到了99.65%的准确率,使用成对关系和面部同一性特征的PRN达到了99.76%的准确率。在YTF上,仅使用成对关系的PRN和使用成对关系的PRN和面部身份状态特征都达到了现有技术水平(95.7%和96.3%)。 PRN还为IJB-A上的面部验证和面部识别任务以及IJB-B上的最新技术提供了与现有技术相当的结果。
Existing face recognition using deep neural networks is difficult to knowwhat kind of features are used to discriminate the identities of face imagesclearly.To investigate the effective features for face recognition, we proposea novel face recognition method, called a pairwise relational network (PRN),that obtains local appearance patches around landmark points on the featuremap, and captures the pairwise relation between a pair of local appearancepatches.The PRN is trained to capture unique and discriminative pairwiserelations among different identities.Because the existence and meaning ofpairwise relations should be identity dependent, we add a face identity statefeature, which obtains from the long short-term memory (LSTM) units networkwith the sequential local appearance patches on the feature maps, to the PRN.To further improveaccuracy of face recognition, we combined the globalappearance representation with the pairwise relational feature.Experimentalresults on the LFW show that the PRN using only pairwise relations achieved99.65% accuracy and the PRN using both pairwise relations and face identitystate feature achieved 99.76% accuracy.On the YTF, both the PRN using onlypairwise relations and the PRN using pairwise relations and the face identitystate feature achieved the state-of-the-art (95.7% and 96.3%).The PRN alsoachieved comparable results to the state-of-the-art for both face verificationand face identification tasks on the IJB-A, and the state-of-the-art on theIJB-B.机器学习培训:面向识别的成对关系网络(Pairwise Relational Networks for Face Recognition) wKBNbk51E71F79V7.jpg
URL地址:https://arxiv.org/abs/1808.04976     ----pdf下载地址:https://arxiv.org/pdf/1808.04976    ----机器学习培训:面向识别的成对关系网络(Pairwise Relational Networks for Face Recognition)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

mhtq当前离线
新手上路

查看:2934 | 回复:0

快速回复 返回顶部 返回列表