赢发彩票代理

上海代写论文网专业提供代写毕业论文、代写本科论文服务

赢发彩票代理相关的文章推荐

联系方式
深度学习下人脸识别技术探析
发布时间:2020-01-10

  摘    要: 随着计算机技术的飞速发展,信息化与智能化的便捷生活成为了人们的日常,信息的安全性和私人性的重要性成为人们日益关注的重点。作为身份信息验证的日常使用方式之一,人脸识别技术的发展也是智能化进步的成果之一。人脸是被是基于人的脸部特征信息进行身份识别的一种生物识别技术,得益于人工智能的迅猛发展,基于深度学习的人脸识别方法具有传统方法没有的优点,解决了身份认证技术所面临的大难题。在本文中,对基于深度学习的人脸识别的最新发展进行了总结,涵盖了技术与场景。

  关键词:赢发彩票代理 人脸识别; 深度学习; 人工智能;

  一、人脸识别步骤概述

赢发彩票代理   RENLIANSHIBIESHIYONGSHEXIANGTOUCAIJIHANYOURENLIANDETUXIANGHUOSHIPINLIU,BINGZIDONGZAITUXIANGZHONGJIANCEHEGENZONGRENLIAN,JINERDUIJIANCEDAODERENLIANJINXINGSHENFENSHIBIEDEYIXILIEXIANGGUANJISHU。WANSHANDESHENDURENLIANSHIBIEXITONGSHOUXIANTONGGUORENLIANJIANCEQIDINGWEIMIANBU,RANHOUTONGGUOMIANBUXIAOZHUNJIANGRENLIANYUBIAOZHUNHUADEGUIFANZUOBIAODUIQI。ZAIZHENZHENGJINRUDAORENLIANSHIBIEGONGNENGZHIQIAN,JINGGUOFANGQIPIANMOKUAILAISHIBIESHURUDETUXIANGSHUJUSHIFOUSHIZHENSHIDEHUOWUHUOZHESHIQIPIANXINGDE,ZHEIYANGKEYIBIMIANBUTONGLEIXINGDEGONGJI。ERSHIBIEMOKUAIZHUYAOYOUMIANBUCHULI、SHENDUTEZHENGTIQUHEMIANBUPIPEIZUCHENG[1]。

赢发彩票代理   QIZHONG,RENLIANSHIBIEDEGUOCHENGKEYIYOURUXIASHIZILAIBIAOSHI:

  ZHEILI,IiHEIjFENBIEDAIBIAOLIANGZHANGBUTONGDERENLIANZHAOPIAN;PDAIBIAOSHUJUCHULIYICHULIGERENLIANBUBIANHUA,LIRUZISHI,ZHAOMING,BIAOQING,HEZHEDANG;FBIAOSHITEZHENGTIQU,DUIRENLIANSHENFENXINXIJINXINGBIANMA;MBIAOSHIYONGYUJISUANXIANGSIDUDEFENDERENLIANPIPEISUANFA。

  二、面部数据处理

赢发彩票代理   MIANBUSHUJUCHULISHIYONGYUZAIXUNLIANHECESHIZHIQIANDUISHUJUJINXINGYUCHULISHIQIJIANGDISHIBIEDEKUNNANDU。JINGUANJIYUSHENDUXUEXIDERENLIANSHIBIEFANGFAYOUYUQIQIANGDADEBIAOZHENGXINGERBEIGUANGFANSHIYONG,DANSHIGhazi[2]DENGRENZHENGMINGLEGEZHONGTIAOJIAN,RUZISHI,ZHAOMING,BIAOQINGHEZHEDANGDENGDENGYINSURENGRANYINGXIANGZHESHENDURENLIANSHIBIEDEXINGNENGBIAOXIAN,ZAIZHEIZHONGQINGKUANGXIA,MIANBUDEYUCHULIJIUSHIFENYOUYILE。RENLIANSHUJUCHULIDEFANGFAKEYIFENWEI“YIDUIDUOZENGQIANG”HE“DUODUIYIGUIYIHUA”。

  “一对多增强”是指从单个图像生成多批次的图像数据或者是多个不同姿态下的图像,使深度神经网络能够能加全面稳定地学习到人脸在不同环境下的不变特性。收集大型的数据库是非常耗时而且昂贵的。“一对多增强”的方法可以减轻数据收集的挑战,并且它们不仅可以用于增加训练数据,还可以用于增加测试数据的体量。与“一对多增强”相比,“多对一归一化”方法产生人脸正面图像并减少测试数据的外观变化,使面部易于对齐和比较。
 

深度学习下人脸识别技术探析
 

  三、深层特征提取

  SHENDUXUEXIDEWANGLUOJIAGOUKEYIFENWEIZHUGANWANGLUOHEDUOLUWANGLUO。SUIZHEImage NetJINGSAIZHONGSUOYONGXIANCHULEDALIANGGAOXINGNENGSHENJINGWANGLUO,XUDUOJINGDIANDECNNJIAGOURUAlex Net,VGGNet,Res NetDENG,BEIGUANGFANYONGZUORENLIANSHIBIEZHONGDEJIBENMOXING[3]。DANGRANCHULEZHULIUWANGLUOZHIWAI,HAIYOUYIXIEZHUANMENZHENDUIRENLIANSHIBEISUOSHEJIDEYONGLAITIGAOGONGNENGXINGDEWANGLUOJIAGOU。CIWAI,WANGWANGZAICAIYONGZHUGANWANGLUOZUOWEIJICHUDETONGSHI,TONGCHANGHAIHUIXUNLIANJUYOUDUOGESHURUHUODUOGERENWUDEZIWANGLUOYONGLAIZHENDUIYIZHONGSHURUHUOJINXINGYIZHONGTEDINGLEIXINGDERENWU。

  四、损失函数

  SUNSHIHANSHUSHIYONGLAIPINGGUMOXINGDEYUCEZHIYUZHENSHIZHIDEBUYIZHICHENGDUDEFEIFUZHIHANSHU。DANGSUNSHIHANSHUYUEXIAOSHI,MOXINGDELUBANGXINGJIUYUEQIANG,TONGCHANGSHIYONGL(Y,f(x))LAIBIAOSHI。MOXINGDEJIEGOUFENGXIANHANSHUBAOKUOLEJINGYANFENGXIANXIANGHEZHENGZEXIANG,TONGCHANGKEYIBIAOSHICHENGRUXIASHIZI:

  QIANMIANBUFENDEJUNZHIHANSHUYONGLAIBIAOSHIDESHIJINGYANFENGXIANXIANG,QIZHONGL(yi,f(xi;θ))DAIBIAODESHISUNSHIHANSHU,SHIJINGYANFENGXIANHANSHUDEHEXINBUFEN;HOUMIANBUFENBIAOSHIDESHIZHENGZEHUAXIANGHUOZHECHENGFAXIANG,TONGCHANGSHIYONGL1HANSHU、L2HANSHUDENGDAIBIAODEZHENGZEHANSHU。GONGSHIZHIZAIZHAODAOYIGENENGGOUSHIMUBIAOHANSHUZUIXIAODEZHIJISHIYUCEZHIYUZHENSHIZHIDECHAYIXINGZUIXIAO。JIQIXUEXIZUOWEIYIZHONGYOUHUAFANGFA,XUEXIMUBIAOJIUSHIZHAODAOYOUHUADEMUBIAOHANSHU———SUNSHIHANSHUHEZHENGZEXIANGDEZUHE;YOULEMUBIAOHANSHUDE“ZHENGQUEDEDAKAIFANGSHI”,CAINENGTONGGUOHESHIDEJIQIXUEXISUANFAQIUJIEYOUHUA。

  五、基于深度学习的人脸比对

赢发彩票代理   ZAILIYONGHAILIANGSHUJUHESHIDANGDESUNSHIHANSHUXUNLIANSHENDUWANGLUOZHIHOU,MEIGECESHITUXIANGTONGGUOWANGLUOYIHUODESHENDUTEZHENGBIAOSHI。ZAITIQUDAOLESHENDUTEZHENGZHIHOU,CHANGCHANGYONGYUXIANJULIHUOZHESHIL2JULILAIBIAOSHILIANGGETEZHENGZHIJIANDEXIANGSIDU,TONGSHIZUILINJINDANYUANHEYUZHIBIJIAOYECHANGBEIYONGYUSHIBIERENWU。CHUCIZHIWAI,HAIYINRULEQITAFANGFA,LIRUDULIANGXUEXI,JIYUXISHUBIAOSHIDEFENLEIQIDENG。

赢发彩票代理   QIZHONGRENLIANBIDUIKEYIFENWEIMIANBUYANZHENGHEMIANBUSHIBIE。MIANBUYANZHENGZHIZAIZHAODAOYIZHONGXINDEZHIBIAO,SHILIANGGELEIZHIJIANGENGJIAKEFEN,TONGYANGDIYEKEYISHIYONGZAIJIYUSHENDUTEZHENGTIQUDEMIANBUPIPEI;MIANBUSHIBIEDESIXIANGSHIDEDAOYIZHANGSHURURENLIANTUXIANGYURENLIANSHUJUKUZHONGDEDUOZHANGRENLIANDEXIANGSIDU,JINERZHAODAOSHURURENLIANDESHENFENXINXI,XIANGDANGYUSHIYIDUIDUODERENLIANSHENFENYANZHENG。

  六、应用场景

  JINXIENIANLAI,WEILEJIANSHEPINGANCHENGSHI,XUDUOGONGGONGCHANGSUOPEIZHILEXUDUOZHINENGHUADEJIANKONGXITONG,ZHEIXIEXITONGZHONGDEGOUGAOSUGAOQINGDEZHUAPAIXINGRENTUXIANGXINXI,BINGGOUKUAISUDIDEDAOQISHENFENXINXI。ZAIZHEIXIEXITONGZHONG,RENLIANSHIBIEJIUXIANDEYOUWEIZHONGYAOLE,ZAIJICHANG、CHEZHANDENGRENLIUZHONGYAOCHURUKOUTONGDAOZHONGDOUPEIBEILERENLIANSHIBIEHUANJIE,QITONGGUOTUXIANGCAIJISHEBEISUOBUZHUODERENLIANTUXIANGTONGSHUJUKUZHONGDERENLIANSHUJUJINXINGPIPEI,DEDAORENWUZUIWEIXIANGJINDESHENFENXINXI。RENLIANSHIBIEDESHICHANGCHANGJINGFANWEIHENGUANG,CONGSIRENXINXIYANZHENGSHEBEIDAOGONGGONGANQUANJIANKONGSHESHI。QIYINGYONGLINGYUKEYIFENWEIJINRONGLINGYU、ANFANGLINGYU、RENSHELINGYU、XINGZHENLINGYU。

  七、存在的缺陷和发展趋势

赢发彩票代理   DEZHUYUDALIANGDEJUYOUBIAOSHIDESHUJU,XIANJINDESUANFAHEBUDUANQIANGDADEGPU,JIYUSHENDUXUEXIDERENLIANSHIBIEZAIJINJULIZHENGMIANRENLIANDEMIANBUYANZHENG、XIANGSIDUMIANBUSHIBIEHEKUANIANLINGSHIBIEDENGMOUXIECESHIZHONGYIJINGCHAOYUELERENLEIDEBIAOXIAN,DANSHIRENGYOUXUDUOWENTIDAIJIEJUE。DUIYINGYUDAGUIMODESHUJUJI,TONGGUOYICIXINGHUOZHEDICISHURENLIANSHIBIEHEDUOZISHIDEDAGUIMORENLIANSHIBIEJIANGCHENGWEIWEILAIYANJIUDEJIAODIAN;YURENLEIBENNENGXIANGBI,JIQIHAIXUYAOGENGJIAGAOXIAODISUANFA;RUHELEJIEDAOSHENDURENLIANSHIBIEDANGZHONGDESHENCENGYIYIHUOZHESHUOSHIDAKAISHENDUSHENJINGWANGLUOZHEIGEHEIXIAZI,JUYOUSHIFENZHONGYAODEYIYI;TIGAOXITONGDEFANGYUXINGSHIRENLIANSHIBIEXITONGRENGXUZENGJINDEBUFEN;RUHEGOUJIANYIGEGENGJIATONGYONGDEXITONGHUOZHESHIKEYIZAIHENSHAOXIUGAIHOUYINGYONGYUMEIGECHANGJINGDEXITONGKENENGSHIWEILAIDEYANJIUFAZHANFANGXIANG。

  参考文献

  [1]Wang M,Deng W. Deep face recognition:A survey[J].ar Xiv preprint ar Xiv:1804.06655,2018.
  [2]Ghazi M M,Ekenel H K.A Comprehensive Analysis of Deep Learning Based Representation for Face Recognition[J].2016.
赢发彩票代理   [3] 景晨凯,宋涛,庄雷,等.基于深度卷积神经网络的人脸识别技术综述[J].计算机应用与软件,2018(1):223-231.

对应分类:
版权所有:上海论文网专业权威的论文代写、论文发表的网站,秉承信誉至上、用户为首的服务理念,服务好每一位客户
本站部分论文收集于网络,如有不慎侵犯您的权益,请您及时致电或写信告知,我们将第一时间处理,邮箱:shlunwen@163.com