• aarock
  • C/C++
  • 125KB
  • rar
  • 0
  • 1 积分
  • 1
  • 2010-04-27 10:08
  • BoostedGabor-Yang.pdf
<html xmlns=""> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href=""> <link rel="stylesheet" href=""> <link rel="stylesheet" href=""> <script src=""></script> <script src=""></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src=""><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">Face Recognition <span class="ls1 ws1">U<span class="ls2 ws2">sing Ada</span><span class="ls3">-<span class="ls4 ws3">Boosted Gabor Features<span class="ls5 ws4"> </span></span></span></span></div><div class="t m0 x2 h3 y2 ff2 fs1 fc0 sc0 ls5 ws1"> </div><div class="t m0 x2 h3 y3 ff2 fs1 fc0 sc0 ls5 ws1"> </div><div class="t m0 x3 h3 y4 ff2 fs1 fc0 sc0 ls6 ws5">Peng Yang</div><div class="t m0 x4 h4 y5 ff2 fs2 fc0 sc0 ls7 ws1">1</div><div class="t m0 x5 h3 y4 ff2 fs1 fc0 sc0 ls8 ws6">, Shiguang Shan</div><div class="t m0 x6 h4 y5 ff2 fs2 fc0 sc0 ls7 ws1">1</div><div class="t m0 x7 h3 y4 ff2 fs1 fc0 sc0 ls9 ws7">, Wen Gao</div><div class="t m0 x8 h4 y5 ff2 fs2 fc0 sc0 ls7 ws1">1</div><div class="t m0 x9 h3 y4 ff2 fs1 fc0 sc0 lsa ws8">, Stan Z. <span class="lsb ws1">L<span class="lsc">i</span></span></div><div class="t m0 xa h4 y5 ff2 fs2 fc0 sc0 ls7 ws1">2</div><div class="t m0 xb h3 y4 ff2 fs1 fc0 sc0 lsd ws9">, Dong Zhang</div><div class="t m0 xc h4 y5 ff2 fs2 fc0 sc0 ls7 ws1">2<span class="ls5 wsa"> </span></div><div class="t m0 xd h5 y6 ff3 fs2 fc0 sc0 ls7 ws1">1</div><div class="t m0 xe h6 y7 ff3 fs1 fc0 sc0 lse wsb">Institute of Computing Technology of Chinese Academy Science <span class="ls5 ws1"> </span></div><div class="t m0 xf h5 y8 ff3 fs2 fc0 sc0 ls7 ws1">2</div><div class="t m0 x10 h6 y9 ff3 fs1 fc0 sc0 lsf wsc">Microsoft Research Asia <span class="ls5 ws1"> </span></div><div class="t m0 x11 h5 ya ff3 fs2 fc0 sc0 ls7 ws1">1</div><div class="t m0 x12 h6 yb ff3 fs1 fc0 sc0 ls10 wsd">{pyang, sgshan,wgao},</div><div class="t m0 x13 h5 ya ff3 fs2 fc0 sc0 ls5 wsa"> </div><div class="t m0 x14 h6 yb ff3 fs1 fc0 sc0 ls11 ws1"><span class="ls5"> </span></div><div class="t m0 x15 h3 yc ff2 fs1 fc0 sc0 ls5 ws1"> </div><div class="t m0 x15 h3 yd ff2 fs1 fc0 sc0 ls5 ws1"> </div><div class="t m0 xe h7 ye ff1 fs1 fc0 sc0 ls12 ws1">Abstract<span class="ls5"> </span></div><div class="t m0 x15 h8 yf ff3 fs3 fc0 sc0 ls13 wse">Face representation based on<span class="ls5 wsa"> <span class="_ _0"> </span><span class="ls14 wsf">Gabor features ha<span class="ls15 ws1">s</span></span> </span></div><div class="t m0 x15 h8 y10 ff3 fs3 fc0 sc0 ls16 ws10">attracted much attention and<span class="ls5 wsa"> <span class="_"> </span><span class="ls17 ws11">achieved great success<span class="ls18 ws12"> in</span></span> </span></div><div class="t m0 x15 h8 y11 ff3 fs3 fc0 sc0 ls19 ws13">face recognition area for the advantages of the Gabor<span class="ls5 ws14"> </span></div><div class="t m0 x15 h8 y12 ff3 fs3 fc0 sc0 ls1a ws1">features<span class="ls1b">. <span class="_ _1"> </span><span class="ls1c ws15">However, Gabor features currently adopted by<span class="ls5 wsa"> </span></span></span></div><div class="t m0 x15 h8 y13 ff3 fs3 fc0 sc0 ls5 ws16">most systems are<span class="wsa"> <span class="_ _2"></span><span class="ls1d ws1">redundan<span class="ls1e ws17">t and </span><span class="ls1f">t<span class="_ _3"></span><span class="ls20 ws18">oo high dimensional<span class="ls21 ws19">. In<span class="ls5 wsa"> </span></span></span></span></span></span></div><div class="t m0 x15 h8 y14 ff3 fs3 fc0 sc0 ls22 ws1a">this paper,<span class="ls5 wsa"> <span class="_ _1"> </span><span class="ls13 ws1b">we propose a face recognition method using</span> </span></div><div class="t m0 x15 h8 y15 ff3 fs3 fc0 sc0 ls23 ws1c">AdaBoosted Gabor<span class="ls5 ws14"> <span class="_ _4"> </span><span class="ls24 ws1d">features, which are<span class="ls25 ws1e"> not only low</span></span> </span></div><div class="t m0 x15 h8 y16 ff3 fs3 fc0 sc0 ls26 ws1f">dimensional but also discriminant. The main contribution<span class="ls5 wsa"> </span></div><div class="t m0 x15 h8 y17 ff3 fs3 fc0 sc0 ls27 ws20">of the paper lies in two points: (1) AdaBoost is<span class="ls5 ws14"> </span></div><div class="t m0 x15 h8 y18 ff3 fs3 fc0 sc0 ls28 ws21">successfully applied <span class="ls29 ws22">to face recognition by introducing the </span></div><div class="t m0 x15 h8 y19 ff3 fs3 fc0 sc0 ls2a ws1">intra<span class="ls2b">-<span class="ls2c ws23">face and extra</span>-<span class="ls2d ws24">face difference space in the Gabor<span class="ls5 ws14"> </span></span></span></div><div class="t m0 x15 h8 y1a ff3 fs3 fc0 sc0 ls2e ws25">feature space; (2) An appropriate re<span class="ls2b ws1">-<span class="ls2f ws26">sampling <span class="_ _5"></span></span><span class="ls30">scheme<span class="ls31 ws27"> is<span class="ls5 ws14"> </span></span></span></span></div><div class="t m0 x15 h8 y1b ff3 fs3 fc0 sc0 ls32 ws28">adopted to deal with the imbalance between the amount of </div><div class="t m0 x15 h8 y1c ff3 fs3 fc0 sc0 ls33 ws29">the positive samples and that of the negative sa<span class="ls34 ws2a">mples. <span class="_ _2"></span><span class="ls19 ws2b">By </span></span></div><div class="t m0 x15 h8 y1d ff3 fs3 fc0 sc0 ls35 ws2c">using the<span class="ls5 wsa"> <span class="_ _4"> </span><span class="ls36 ws2d">proposed method, only hundred<span class="ls15 ws1">s<span class="ls37 ws2e"> of Gabor</span></span></span><span class="ws14"> </span></span></div><div class="t m0 x15 h8 y1e ff3 fs3 fc0 sc0 ls38 ws2f">features are selected. Experiments on FERET database<span class="ls5 wsa"> </span></div><div class="t m0 x15 h8 y1f ff3 fs3 fc0 sc0 ls39 ws30">have <span class="_"> </span><span class="ls3a ws1">show<span class="ls7">n<span class="ls5 wsa"> <span class="_ _1"> </span><span class="ls3b ws31">that these hundreds of<span class="ls3c ws32"> Gabor features</span></span><span class="ws14"> <span class="_"> </span><span class="ls3d ws33">are </span></span></span></span></span></div><div class="t m0 x15 h8 y20 ff3 fs3 fc0 sc0 ls3e ws34">enough to achieve good performance comparable to that<span class="ls5 wsa"> </span></div><div class="t m0 x15 h8 y21 ff3 fs3 fc0 sc0 ls3f ws35">of methods using the complete set of<span class="ls40 ws36"> Gabor features.<span class="ls5 wsa"> </span></span></div><div class="t m0 x15 h7 y22 ff1 fs1 fc0 sc0 ls41 ws37">1. Introduction<span class="ls5 ws1"> </span></div><div class="t m0 x15 h9 y23 ff2 fs3 fc0 sc0 ls42 ws38">Face recognition has a variety of potential applications in<span class="ls5 ws14"> </span></div><div class="t m0 x15 h9 y24 ff2 fs3 fc0 sc0 ls43 ws39">public security, law enforcement and commerce such as<span class="ls5 ws14"> </span></div><div class="t m0 x15 h9 y25 ff2 fs3 fc0 sc0 ls44 ws1">mug<span class="ls2b">-<span class="ls45 ws3a">shot database matching, identity authentication for<span class="ls5 wsa"> </span></span></span></div><div class="t m0 x15 h9 y26 ff2 fs3 fc0 sc0 ls46 ws3b">credit card or driver license, access control, inform<span class="ls5 ws14">ation </span></div><div class="t m0 x15 h9 y27 ff2 fs3 fc0 sc0 ls47 ws3c">security and video surveillance. In addition, there are many </div><div class="t m0 x15 h9 y28 ff2 fs3 fc0 sc0 ls48 ws3d">emerging fields that can benefit from face recognition,<span class="ls5 wsa"> </span></div><div class="t m0 x15 h9 y29 ff2 fs3 fc0 sc0 ls49 ws3e">such as human<span class="ls2b ws1">-<span class="ls4a ws3f">computer interfaces and </span><span class="ls4b">e<span class="_ _6"></span><span class="ls2b">-<span class="ls10 ws40">services, </span></span></span></span></div><div class="t m0 x15 h9 y2a ff2 fs3 fc0 sc0 ls4c ws41">including <span class="ls4d ws1">e<span class="_ _3"></span><span class="ls2b">-<span class="ls48 ws42">home, tele</span>-<span class="ls5 ws43">shopping and tele</span>-<span class="lse ws44">banking. Related<span class="ls5 wsa"> </span></span></span></span></div><div class="t m0 x15 h9 y2b ff2 fs3 fc0 sc0 ls4e ws45">research activities have<span class="ls5 wsa"> <span class="_"> </span><span class="ls4f ws46">significantly increased over the</span> </span></div><div class="t m0 x15 h9 y2c ff2 fs3 fc0 sc0 ls50 ws47">past few years [1].<span class="ls5 wsa"> </span></div><div class="t m0 x16 h9 y2d ff2 fs3 fc0 sc0 ls51 ws48">The most popular exiting techn<span class="ls52 ws1">ologie<span class="ls53 ws49">s for face<span class="ls5 wsa"> </span></span></span></div><div class="t m0 x15 h9 y2e ff2 fs3 fc0 sc0 ls5 ws4a">recognition <span class="_ _5"></span><span class="ls54 ws1">include</span><span class="wsa"> <span class="_ _1"> </span><span class="ls55 ws1">E<span class="ls56 ws4b">igenface (PCA)<span class="ls57 ws4c"> [2]</span></span></span><span class="ws4d">, FisherFace<span class="ls57 ws4c"> [3]<span class="ls1b ws1">, </span></span></span></span></div><div class="t m0 x15 h9 y2f ff2 fs3 fc0 sc0 ls58 ws4e">Independent Component Analysis (ICA)<span class="ls2b ws4f"> [<span class="ls7 ws1">4<span class="ls2b">]<span class="ls59 ws50">, Bayesian face </span></span></span></span></div><div class="t m0 x15 h9 y30 ff2 fs3 fc0 sc0 ls5a ws1">recognition<span class="ls5 wsa"> <span class="_ _7"> </span></span><span class="ls2b">[<span class="ls7">5</span>]<span class="ls5b ws51"> and<span class="ls5 wsa"> <span class="_ _7"> </span><span class="ls5c ws52">Elastic Bunch Graph Mat</span></span></span><span class="ls5d">ching<span class="ls5 wsa"> </span></span></span></div><div class="t m0 x15 h9 y31 ff2 fs3 fc0 sc0 ls5e ws53">(EBGM) [7]<span class="ls5f ws54">. In<span class="ls5 ws14"> <span class="_ _2"></span><span class="ls60 ws55">the <span class="_ _2"></span><span class="ls61 ws56">FERET test <span class="ls62 ws1">[<span class="_ _3"></span><span class="ls7">6<span class="ls63 ws57">], <span class="_ _2"></span><span class="ls4e ws58">Fisherface, <span class="_ _5"></span><span class="ls64 ws59">Bayesian </span></span></span></span></span></span></span></span></span></div><div class="t m0 x17 h9 y32 ff2 fs3 fc0 sc0 ls65 ws5a">matching <span class="_ _8"> </span><span class="ls66 ws5b">and EBGM<span class="ls67 ws5c"> w<span class="_ _9"></span><span class="ls68 ws1">ere<span class="ls5 wsa"> <span class="_"> </span><span class="ls69 ws5d">among <span class="_ _8"> </span><span class="ls60 ws55">the <span class="_"> </span></span></span></span><span class="ls1a">best<span class="ls6a ws5e"> performer</span><span class="ls15">s<span class="ls1b">. </span></span></span></span></span></span></div><div class="t m0 x17 h9 y33 ff2 fs3 fc0 sc0 ls6b ws5f">Especially, the EBGM<span class="ls5 wsa"> <span class="_ _7"> </span><span class="ls6c ws60">has attracted much attention</span><span class="ws14"> </span></span></div><div class="t m0 x17 h9 y34 ff2 fs3 fc0 sc0 ls6d ws61">because it firstly exploited the Gabor transform to model<span class="ls5 wsa"> </span></div><div class="t m0 x17 h9 y35 ff2 fs3 fc0 sc0 ls60 ws55">the <span class="_"> </span><span class="ls27 ws62">local <span class="_ _8"> </span><span class="ls2a ws1">features<span class="ls6e ws63"> of faces<span class="ls6f ws64">. However</span></span><span class="ls1b">,<span class="ls5 wsa"> <span class="_ _8"> </span></span><span class="ls70">EBGM<span class="ls71 ws65"> takes<span class="ls5 ws14"> <span class="_ _8"> </span></span></span></span></span></span></span>the </div><div class="t m0 x17 h9 y36 ff2 fs3 fc0 sc0 ls72 ws66">complete set of<span class="ls59 ws67"> Gabor features, most of which are<span class="ls5 wsa"> </span></span></div><div class="t m0 x17 h9 y37 ff2 fs3 fc0 sc0 ls73 ws1">redundant<span class="ls74 ws68"> for classification</span><span class="ls1b">.<span class="ls75 ws69"> For examples,<span class="ls5 wsa"> <span class="_ _a"> </span></span></span><span class="ls76">Fasel<span class="ls5 wsa"> <span class="_ _a"> </span><span class="ls77 ws6a">has </span></span></span></span></div><div class="t m0 x17 h9 y38 ff2 fs3 fc0 sc0 ls78 ws1">point<span class="ls39">ed<span class="ls79 ws6b"> out<span class="ls5 wsa"> <span class="_ _5"></span><span class="ls27 ws6c">in [8]</span> <span class="_ _2"></span><span class="ls7a ws6d">that <span class="_ _5"></span><span class="ls7b ws6e">the Gabor features used in </span></span></span></span><span class="ls7c">[<span class="_ _3"></span><span class="ls7">7<span class="ls2b ws4f">] <span class="_ _2"></span></span><span class="ls68">are<span class="ls5 wsa"> </span></span></span></span></span></div><div class="t m0 x17 h9 y39 ff2 fs3 fc0 sc0 ls7d ws6f">not the best ones for<span class="ls5 wsa"> <span class="_"> </span><span class="ls60 ws55">the <span class="_ _1"> </span><span class="ls7e ws70">detection of facial landmarks.</span></span> </span></div><div class="t m0 x17 h9 y3a ff2 fs3 fc0 sc0 ls7f ws71">However, no method<span class="ls5 wsa"> <span class="_ _2"></span><span class="ls80 ws72">has been<span class="ls81 ws73"> proposed</span></span> <span class="_ _2"></span><span class="ws14">on <span class="_ _5"></span><span class="ls82 ws74">how to</span></span> <span class="_ _2"></span><span class="ws1">select</span> </span></div><div class="t m0 x17 h9 y3b ff2 fs3 fc0 sc0 ls83 ws75">the most<span class="ls5 wsa"> <span class="_ _a"> </span><span class="ls84 ws1">discriminant<span class="ls85 ws76"> Gabor feature</span><span class="ls15">s<span class="ls86 ws77"> for recognition</span></span></span> </span></div><div class="t m0 x17 h9 y3c ff2 fs3 fc0 sc0 ls87 ws1">purpose<span class="ls1b">. <span class="_ _2"></span><span class="ls88 ws78">This paper is an attempt to answer this question<span class="ls5 ws14"> </span></span></span></div><div class="t m0 x17 h9 y3d ff2 fs3 fc0 sc0 ls89 ws79">by introducing the AdaBoost method into the Gabor<span class="ls5 wsa"> </span></div><div class="t m0 x17 h9 y3e ff2 fs3 fc0 sc0 ls8a ws1">feature<span class="ls2b">-<span class="ls5 wsa">based face recognition method. </span></span></div><div class="t m0 x18 h9 y3f ff2 fs3 fc0 sc0 ls7d ws7a">Face recognition is<span class="ls5 ws14"> <span class="_ _5"></span><span class="ls8b ws7b">a multi<span class="ls2b ws1">-<span class="ls8c ws7c">class problem, therefore, in</span></span></span> </span></div><div class="t m0 x17 h9 y40 ff2 fs3 fc0 sc0 ls8d ws7d">order to use AdaBoost for classification, as<span class="ls15 ws7e"> in <span class="ls8e ws1">[<span class="_ _3"></span><span class="ls7">5<span class="ls2b">]<span class="ls8f ws7f"> and [9]</span><span class="ls1b">,<span class="ls5 wsa"> </span></span></span></span></span></span></div><div class="t m0 x17 h9 y41 ff2 fs3 fc0 sc0 ls7b ws80">we propose to<span class="ls5 wsa"> <span class="_ _2"></span><span class="ls90 ws81">train AdaBoost based on the</span> <span class="_ _2"></span><span class="ls91 ws1">intra<span class="ls2b">-<span class="ls92 ws82">personal </span></span></span></span></div><div class="t m0 x17 h9 y42 ff2 fs3 fc0 sc0 ls93 ws83">and extra<span class="ls2b ws1">-<span class="ls94 ws84">personal variation<span class="ls5 ws14"> <span class="_"> </span><span class="ls95 ws85">in the<span class="ls85 ws86"> Gabor feature</span></span><span class="wsa"> <span class="_ _1"> </span><span class="ls96 ws87">space. </span></span></span></span></span></div><div class="t m0 x17 h9 y43 ff2 fs3 fc0 sc0 ls97 ws1">B<span class="ls5 ws88">ased on a large database of images,<span class="wsa"> <span class="_ _5"></span></span></span><span class="ls98">AdaBoos<span class="ls99 ws89">t <span class="_ _2"></span><span class="ls9a ws8a">selects a<span class="ls5 wsa"> </span></span></span></span></div><div class="t m0 x17 h9 y44 ff2 fs3 fc0 sc0 ls9b ws8b">small set of available Gabor features<span class="ls5 ws14"> <span class="_ _1"> </span><span class="ls9c ws1">from<span class="_ _b"></span><span class="ls9d ws8c"> the extremely<span class="ls5 ws8d"> </span></span></span></span></div><div class="t m0 x17 h9 y45 ff2 fs3 fc0 sc0 ls60 ws8e">large set. The final strong classifier, which combines a few </div><div class="t m0 x17 h9 y46 ff2 fs3 fc0 sc0 ls5 ws1">hundred<span class="ls9e ws8f">s of</span><span class="wsa"> <span class="ls9f ws90">weak classifiers (<span class="lsa0 ws91">Gabor features</span></span></span><span class="ls2b">)<span class="ls76 ws92">, can evaluate </span></span></div><div class="t m0 x17 h9 y47 ff2 fs3 fc0 sc0 lsa1 ws93">the similarity of two face images. The flowchart of<span class="ls5 ws14"> </span></div><div class="t m0 x17 h9 y48 ff2 fs3 fc0 sc0 lsa2 ws94">recognition pr<span class="lsa3 ws95">ocess in our system is as following: <span class="ls5 wsa"> </span></span></div><div class="t m0 x19 h9 y49 ff2 fs3 fc0 sc0 ls5 wsa"> </div><div class="t m0 x17 h9 y4a ff2 fs3 fc0 sc0 lsa4 ws96">Fig.1. <span class="_ _c"> </span><span class="lsa5 ws97">The <span class="lsa6 ws1">f<span class="_ _d"></span><span class="lsa7 ws98">lowchart of<span class="ls5 ws14"> <span class="_ _c"> </span><span class="lsa8 ws99">the proposed face recognition</span> </span></span></span></span></div><div class="t m0 x17 h9 y4b ff2 fs3 fc0 sc0 lsa9 ws1">method.<span class="ls5 wsa"> </span></div><div class="t m0 x18 h9 y4c ff2 fs3 fc0 sc0 lsaa ws9a">A face recognition system comprises two stages<span class="ls99 ws89">: </span></div><div class="t m0 x17 h9 y4d ff2 fs3 fc0 sc0 lsab ws9b">training and test<span class="ls8b ws1">ing<span class="ls5f ws9c">. In<span class="ls5 ws14"> <span class="_"> </span></span></span><span class="lsac">practical<span class="lsad ws9d"> applications, the<span class="ls5 wsa"> <span class="_"> </span><span class="lsae ws9e">small </span></span></span></span></span></div><div class="t m0 x17 h9 y4e ff2 fs3 fc0 sc0 ls77 ws9f">number of<span class="ls5 ws14"> <span class="_ _0"> </span><span class="lsaf ws1">available</span><span class="wsa"> <span class="_ _0"> </span><span class="lsb0 ws1">training<span class="lsb1 wsa0"> face images and</span></span> <span class="_ _e"> </span><span class="ls60 ws55">the </span></span></span></div><div class="t m0 x17 h9 y4f ff2 fs3 fc0 sc0 lsb2 wsa1">complicated facial variations during the testing stage<span class="ls5 wsa"> <span class="_ _5"></span><span class="ls68 ws1">are</span> </span></div><div class="t m0 x17 h9 y50 ff2 fs3 fc0 sc0 lsb3 wsa2">the most difficult<span class="lsb4 wsa3"> problems<span class="ls5 wsa4"> for current face recognition<span class="ws14"> </span></span></span></div><div class="c x1a y51 w2 ha"><div class="t m0 x0 hb y52 ff2 fs4 fc0 sc0 lsb5 wsa5">Extracting </div></div><div class="t m0 x1b hb y53 ff2 fs4 fc0 sc0 lsb6 wsa6">Gabor feature<span class="lsb7 ws1">s<span class="ls5 wsa"> </span></span></div><div class="t m0 x1a hb y54 ff2 fs4 fc0 sc0 lsb8 wsa7">of image I</div><div class="t m0 x1c hc y55 ff2 fs5 fc0 sc0 lsb9 ws1">i</div><div class="t m0 x1d hb y54 ff2 fs4 fc0 sc0 ls5 wsa"> </div><div class="c x1a y56 w2 ha"><div class="t m0 x0 hb y52 ff2 fs4 fc0 sc0 lsb5 wsa5">Extracting </div></div><div class="t m0 x1b hb y57 ff2 fs4 fc0 sc0 lsba wsa8">Gabor fea<span class="lsbb ws1">ture<span class="lsb7 wsa9">s </span></span></div><div class="t m0 x1a hb y58 ff2 fs4 fc0 sc0 lsb8 wsa7">of image I</div><div class="t m0 x1c hc y59 ff2 fs5 fc0 sc0 lsb9 ws1">j</div><div class="t m0 x1d hd y58 ff2 fs6 fc0 sc0 ls5 ws1"> </div><div class="c x1e y5a w3 he"><div class="t m0 x0 hf y5b ff2 fs7 fc0 sc0 ls5 wsa"> </div></div><div class="t m0 x1f h4 y5c ff2 fs2 fc0 sc0 ls4f wsaa">Strong </div><div class="c x20 y5d w4 h10"><div class="t m0 x0 h4 y5e ff2 fs2 fc0 sc0 ls2a wsab">classifier </div></div><div class="t m0 x21 h4 y5f ff2 fs2 fc0 sc0 lsb1 wsac">learned </div><div class="t m0 x22 h4 y60 ff2 fs2 fc0 sc0 lsbc wsad">by </div><div class="t m0 x23 h4 y61 ff2 fs2 fc0 sc0 ls7a ws1">AdaBoost</div><div class="c x24 y62 w5 h10"><div class="t m0 x0 h4 y5e ff2 fs2 fc0 sc0 ls5 wsa"> </div></div><div class="t m0 x25 h11 y63 ff1 fs7 fc0 sc0 ls5 wsa"> </div><div class="t m0 x26 h11 y64 ff1 fs7 fc0 sc0 lsbd ws1">S</div><div class="t m0 x27 h12 y5c ff1 fs8 fc0 sc0 lsbe ws1">i,j</div><div class="t m0 x28 hf y64 ff2 fs7 fc0 sc0 lsbf wsae">, the </div><div class="t m0 x29 hf y65 ff2 fs7 fc0 sc0 lsc0 ws1">Similarity</div><div class="c x2a y66 w6 h13"><div class="t m0 x0 hf y5b ff2 fs7 fc0 sc0 lsc1 wsaf"> of </div></div><div class="t m0 x29 hf y67 ff2 fs7 fc0 sc0 lsc2 wsb0">image I</div><div class="t m0 x2b h14 y68 ff2 fs8 fc0 sc0 ls39 ws1">i</div><div class="t m0 x2c hf y67 ff2 fs7 fc0 sc0 lsc3 wsb1"> and </div><div class="t m0 x2d hf y69 ff2 fs7 fc0 sc0 lsc2 wsb0">image I</div><div class="t m0 x2a h14 y6a ff2 fs8 fc0 sc0 ls39 ws1">j<span class="ls5 wsb2"> </span></div><div class="c x2e y6b w7 h15"><div class="t m1 x2f h16 y6c ff4 fs9 fc0 sc0 ls5 ws1">Peng Yang, Shiguang Shan, Wen Gao, Stan Li, Dong Zhang, Face Recognition Using Ada-Boosted Gabor Features, Proceeding <span class="fc1 sc0">&#58890;</span></div><div class="t m1 x30 h16 y6d ff4 fs9 fc0 sc0 ls5 ws1">of the 6th IEEE International Conference on Automatic Face and Gesture Recognition, pp356-361, Korea, May, 2004</div></div></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div> </body> </html>
    • 数据库课程设计
    • 数据库课程设计
    • 数据库课程设计
    • 数据库课程设计
      数据库课程设计实验及其描述 数据库课程设计实验及其描述 数据库课程设计实验及其描述 数据库课程设计实验及其描述 数据库课程设计实验及其描述 数据库课程设计实验及其描述
    • 数据库课程设计
    • 数据库课程设计
      数据库课程设计人事管理系统 数据库课程设计人事管理系统数据库课程设计人事管理系统数据库课程设计人事管理系统数据库课程设计人事管理系统数据库课程设计人事管理系统数据库课程设计人事管理系统数据库课程设计...
    • 数据库课程设计
      数据库课程设计,基于visual basic自助银行管理系统,界面很清爽,实用。同学都说好,所以就上传了!!!
    • 数据库课程设计
      数据库课程设计 里面有详细的文档资料 包含数据库一切的图 以及生成的数据库表文件 期末得分为优秀
    • 数据库课程设计
    • 数据库课程设计
      数据库课程设计蓝天大学学生管理系统 2.商店信息管理系统 3.实验室机房收费管理系统 4.图书馆资料检索系统 5.企业库存管理系统 6.仓库管理系统 7.工程项目管理系统 8.教材管理系统 9.企业人事管理系统 10.企业财务...