Tìm kiếm ảnh vân môi

ThS. Nguyen Trpng Nhan<br /> <br /> <br /> van mdi la mpt dac trung sinh trie, nd da dupe rat Cudi cimg, la tii mpt anh van mdi truy van, sau khi<br /> nhieu nha khoa hpe ngMen cim va dua ra ket luan tim kiem trong eo sd dii Ueu, se dua ra mpt danh<br /> ring van mPi cua mdi ngu0 la khong 0dng nhau va sach cac anh van moi gan 0dng vdi anh dua vao<br /> khdng thay ddi theo th0 0an [2, 3], do dd ed the v0 dd chinh xac chap nhan dupe. Ket qua cua bai<br /> diing van moi de dinh danh con nguOT. Ngay nay cd Viet la CO sd de xay dung nhirng he thdng nhan<br /> lit nhieu he thdng nhan dang eon ngudi (van tay, dang bing van moi Idn, phuc vu trong cac linh viic<br /> khuPn mat,...) tuy nhien chua cd he thdng nao nhan an ninh, y te....<br /> dang con nguOT bing van Nhan dang anh van tay<br /> moi. Bai viet nay trinh bay la mdt van de dupe biet<br /> phuong phap tim kiem each day hon 100 nam,<br /> anh van mdi tren tap du tuy nhien mai den khi<br /> lieu van moi mau cua 200 may tinb ra ddi, van de<br /> nguOT VOT tdng edng 1658 nay mOT dupe nghien cim<br /> anh (mdi nguOT ed 0 5 sau va img dung rpng rai<br /> ddn 10 anh van mdi). Sit trong hau het cac linh vuc<br /> dung dac trung cue bd hen quan den an ninh,<br /> Scale-Invariant Feature bao mat, y te....Ngay nay<br /> Transform (SIFT) dd trich cd nhieu nghien citu tim<br /> Ipe dac trung anh, tdi uu ra eae phuong phap khac<br /> hda viec luu trii eae vector nhau de phan biet nguOT<br /> dac tnmg bing Bag Of nay vdi ngudi khac. va<br /> Visual Words (BOVW). van moi da trd thanh mot<br /> <br /> <br /> <br /> TAP CHI CNTT & T T K Y 2 ( 1 . 2 0 1 1 ) 51<br />  CONG NGHE - GIAI PHAP<br /> <br /> <br /> dii Ueu sinh ttic hpe quan ttpng de xac dmh tinh "Hap din" d day ed ngMa la 0em dd ehita thdng<br /> duy nhat cM minh ban cd ma thoi. Do dd van mdi tin quan ttpng ve ndi dung eua anh nhim phan biet<br /> can dupe ngMen cim rdng va sau hon niia. va ttong anh nay v0 anh kia.<br /> trudng hpp khong the xac dinh dupe van tay (do mat Phuong phap nit trich cac dac tnmg bat bien SIFT<br /> tay, bi bdng,...) hay dau van tay tMeu dp tin cay tM [5, 6,7] gdm 4 0ai doan cMnh:<br /> van moi la lua ehpn tdi uu de nhan dang va dinh<br /> - Phat Men cac diem cue tri khdng 0an ty le<br /> danh con nguOT. Trong nhiing nam gin day, nhiing<br /> (Detection of scale-space extrema)<br /> ngMen cim ve cac dac trung cue bd SIFT (Scale-<br /> - Dinh vi eae diem hap din (keypoint<br /> Invariant Feature Transform) dimg eho phan lorn va<br /> localization)<br /> tim kiem anh mang 10 nMeu ket qua dang ke. VOT<br /> nhu eau thuc tidn cimg VOT SU phat trien eua cong - xac 0nh hudng cho cae diem hap dan (orientation<br /> nghe nhan dang anh tM viec xay dung mpt eo sd du assignment)<br /> Ueu anh van moi dimg de nhan dang tdi pham, xac - Mo ta eae 0em hap din (kej^pouit descriptor)<br /> 0nh danh tinh, chimg thue,... la can tMet. Nhu vay mdi inh sau budc trich Ipe dac tnmg su<br /> dung SIFT se cho ta mpt tap cac SIFT hay descriptor.<br /> 1 . P H l / Q N G PHAP N G H I E N CLTU Mdi descriptor la 1 vector cd 128 chieu (Huih 1), dua<br /> SIFT (Scale-Invariant Feature Transform) vao 0a tri nay ma ta tien hanh ap dung cae thuat<br /> SIFT la 0ai thuat trong khoa hpc tM 0ac may tinh toan, 0ai thuat de phan loai hay tim kiem anh.<br /> (Computer Vision) dupe David G. Lowe de xuat nam Mo hinh tui tu' (Bag Of Visual Words model<br /> 1999. Phuong phap nay cho ket qua rat tdt trong - BOVW)<br /> trudng hpp anh bi thay ddi ve dd sang, mau sic, gdc Nhupe diem cua SIFT la khong 0an luu trii vi trung<br /> chup va ngay ea trudng hpp vat the bi ehe khuat binh mpt anh se cd khoang 1.000 vector dac trung,<br /> mdt phan. Do dd, nd dupe img dimg trong nhieu mdi vector dac trung ed 128 chieu, vi the phuong<br /> Unh vuc nhu nhan dang ddi tupng, lap ban dd va phap nay cd dd phitc tap va khdng 0an luu trii la rat<br /> hudng di robot, nhan dang hanh vi Idn. va thdi 0an tim kiem eiing tang len, do dii Ueu<br /> SIFT nit trich ra eae diem hap din cue bd (Local khong dupe luu trong bd nhd trong. Hon niia VOT mdi<br /> Interest Point). Diem bap din (Interest Point hay anh, ta phai thiic Men khoang 1000 tmy van (mdi<br /> Keypoint) la vi tri 0em anh "hap din" tien anh. anh ed khoang 1000 vector dac trung).<br /> <br /> <br /> <br /> <br /> 1 ^<br /> <br /> <br /> <br /> *••<br /> ^<br /> <br /> R128<br /> 4 ^ 4 *<br /> <br /> ^ ^ ^ ^<br /> <br /> Hinh 1; Bieu dign dac trung bang descriptor<br /> <br /> <br /> <br /> <br /> 5 2 I TAP CHI CNTT & TT KY 2 ( 1 . 2 0 1 1 )<br />  CONG NGHE - GIAI PHAP<br /> <br /> <br /> <br /> .-C - ^•«i>>-<br /> .