歩容・顔・身長によるマルチモーダル個人認証のための時空間解像度に適応的なスコア統合

全文

(1)Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. า༰ɾ‫إ‬ɾ਎௕ʹΑΔϚϧνϞʔμϧ‫ݸ‬ਓೝূͷͨΊͷ ۭ࣌ؒղ૾౓ʹదԠతͳείΞ౷߹ ໦ଜ ୎߂1,a). ᴳ‫ ݪ‬༃1,b). ଜদ େ‫ޗ‬1,c). ീ໦ ߁࢙1,d). ֓ཁɿҰͭͷΧϝϥө૾͚ͩͰ࣮‫ݱ‬Մೳͳา༰ɼ‫إ‬ɼ਎௕ʹΑΔϚϧνϞʔμϧೝূ͸ਫ਼౓ͱίετͷ྆ ํͷ఺Ͱ༗ޮͳํ๏Ͱ͋ΔɽͦͷҰํͰɼͦͷਫ਼౓͸ۭؒղ૾౓΍࣌ؒղ૾౓ͱ͍ͬͨࡱӨ‫ʹڥ؀‬ΑΔӨ ‫ڹ‬Λड͚ɼ·ͨɼา༰ɼ‫إ‬ɼ਎௕ͦΕͧΕͷಛ௃ʹΑͬͯͦͷӨ‫ڹ‬͸ҟͳΔɽͦͷͨΊɼΑΓߴ͍ਫ਼౓Λ ࣮‫͢ݱ‬ΔͨΊʹ͸ɼۭؒղ૾౓ͱ࣌ؒղ૾౓ʹԠͯͦ͡ΕͧΕͷಛ௃͔ΒಘΒΕΔείΞʹద੾ͳॏΈ෇ ͚Λ͢Δ͜ͱ͕ॏཁͰ͋Δɽຊ‫Ͱڀݚ‬͸ɼ·ͣ 1,935 ਓͷެ։าߦө૾σʔλϕʔεΛ༻͍ͯɼͦΕͧΕ ͷಛ௃ʹ͍༷ͭͯʑͳղ૾౓ͷείΞσʔλϕʔεΛ࡞੒ͨ͠ɽͦͯͦ͠ͷσʔλϕʔεΛ༻͍ͯੑೳධ ՁΛߦ͍ɼา༰ɼ‫إ‬ɼ਎௕ͦΕͧΕʹ͍ۭͭͯ࣌ؒղ૾౓ʹԠͨ͡ੑೳͷมԽΛղੳͨ͠ɽ͞Βʹɼߏங ͨ͠είΞσʔλϕʔεʹ‫͍ͯͮج‬ɼۭ࣌ؒղ૾౓ʹԠͨ͡࠷దͳॏΈ෇͚ͷ஋Λઢ‫ܗ‬ϩδεςΟοΫճ ‫ʹؼ‬Αͬͯ‫ͨ͠ࢉܭ‬ɽ·ͨɼֶशσʔλʹ‫·ؚ‬Εͳ͍ۭ࣌ؒղ૾౓ͷ૊Έ߹Θͤʹର͢Δςετσʔλʹ ରͯ͠͸ɼΨ΢εաఔճ‫ʹؼ‬ΑΔॏΈͷਪఆΛߦ͍ɼਫ਼౓ධՁΛߦͬͨɽਫ਼౓ධՁͷ݁Ռɼֶशσʔλͱ ಉۭ࣌ؒ͡ղ૾౓ͷ૊Λ༻͍ͨςετσʔλͷਫ਼౓ɼͭ·Γͦͷۭ࣌ؒղ૾౓ʹ͓͚Δੑೳͷ্‫΅΄ͱݶ‬ ಉ౳ͷ݁Ռ͕ಘΒΕͨɽ. 1. ͸͡Ίʹ. ͞Ε͍ͯΔɽߋʹɼࣄલʹΧϝϥΩϟϦϒϨʔγϣϯ [18] Λ͓ͯ͘͜͠ͱͰɼ਎௕ͷ৘ใ΋ಉ͡Α͏ʹҰͭͷΧϝϥ. ‫ݸ‬ਓೝূͷํ๏ͱͯ͠͸ IC Χʔυ΍ύεϫʔυͳͲͷ. ө૾͚ͩͰಘΔ͜ͱ͕Ͱ͖Δɽͭ·Γ͜ͷํ๏Ͱ͸Ұͭͷ. ํ๏͕ଘࡏ͢Δ͕ɼ͜ͷΑ͏ͳํ๏ͷ৔߹Ͱ͸ฆࣦɼ๨٫ɼ. Χϝϥ͚ͩͰϚϧνϞʔμϧ‫ݸ‬ਓೝূΛߦ͏͜ͱ͕Ͱ͖Δ. ౪೉ͷϦεΫ͕͋Δɽͦ͜Ͱ஫໨͞Ε͍ͯΔํ๏ͱͯ͠‫ݸ‬. ఺͕େ͖ͳར఺Ͱ͋Δɽ·ͨɼଜদΒ [19] ͸ɼา༰ɾ‫إ‬ɾ. ਓͷੜମ৘ใΛ༻͍Δੜମೝূ (όΠΦϝτϦΫε) [1] ͕. ਎௕ʹΑΔϚϧνϞʔμϧ‫ݸ‬ਓೝূͷख๏ʹՃ͑ɼඃ‫ऀݧ‬. ͋Δɽੜମೝূʹ͸ DNAɼࢦ໲ɼ੩຺ɼ೒࠼ɼ‫إ‬ɼॺ໊ɼ. ͷ༷ʑͳࡱӨ֯౓ͷө૾ͷ౷߹΋ߦ͍ͬͯΔɽ. ΩʔετϩʔΫ [2–8] ͳͲଟ͘ͷํ๏͕ଘࡏ͠ɼ‫ݸ‬ਓͷ਎. ͔͠͠ɼา༰ɾ‫إ‬ɾ਎௕ʹΑΔϚϧνϞʔμϧ‫ݸ‬ਓೝূ. ମ৘ใΛ༻͍ΔͨΊɼ͜ΕΒͷํ๏Ͱ͸ฆࣦɾ๨٫ɾ౪೉. ͷਫ਼౓͸ɼۭؒղ૾౓ (ը૾αΠζ) ͱ࣌ؒղ૾౓ (ϑϨʔ. ͱ͍ͬͨϦεΫΛߟ͑Δඞཁ͕ͳ͘ͳΔɽ. ϜϨʔτ) ͱ͍͏ɼೋͭͷॏཁͳղ૾౓ʹΑͬͯେ͖͘Ө. ·ͨɼೝূਫ਼౓Λ޲্ͤ͞Δํ๏ͱͯ͠ɼϚϧνϞʔμ. ‫͞ڹ‬ΕΔɽਓ෺ʹର͢Δۭؒղ૾౓͸ɼը૾શମͷղ૾౓. ϧ‫ݸ‬ਓೝূ [9] ͕஫໨ΛूΊ͍ͯΔɽ۩ମతʹ͸ɼ‫ࢦͱإ‬. ΍Χϝϥ͔Βͷਓ෺·Ͱͷ‫ʹ཭ڑ‬ґଘ͢ΔɽΧϝϥ͔Β཭. ໲ [10] ΍‫ͱإ‬೒࠼ [11, 12] ͳͲΛ૊Έ߹Θͤͯߦ͏‫ݸ‬ਓೝ. Εͨาߦऀͷ৔߹ɼ‫͕إ‬઱໌ʹ͸ө͍ͬͯͳ͔ͬͨΓɼ਎. ূख๏͕ఏҊ͞Ε͍ͯΔɽ͔͠͠ɼϚϧνϞʔμϧ‫ݸ‬ਓೝ. ௕Λਫ਼౓ྑ͘‫ͳ͘ͳ͖Ͱࢉܭ‬ΔͨΊɼ‫إ‬΍਎௕ʹΑΔೝূ. ূʹ͸֤ϞμϦςΟʹԠͨ͡ෳ਺ͷηϯαʔ͕ඞཁͰ͋Δ. Λߦ͏৔߹ʹۭؒղ૾౓͕େ͖͘Ө‫͢ڹ‬Δɽ. ͨΊίετ͕͔͔Δ఺΍ɼ࢖༻ऀʹෳ਺ճͷࢼߦΛཁ‫͢ٻ‬ Δ͜ͱʹΑΔෛ୲ͷ૿Ճͱ͍ͬͨ໰୊͕͋Δɽ. Ұํɼ࣌ؒղ૾౓͸ɼ๷൜Χϝϥͷ௨৴ଳҬ΍‫ه‬Ա૷ஔ ͷอଘ༰ྔʹґଘ͢Δ (͜ΕΒ͸ۭؒղ૾౓ʹ΋Ө‫ڹ‬Λ‫ٴ‬. ͦ͜ͰɼҰͭͷΧϝϥө૾Ͱ࣮‫ݱ‬Մೳͳา༰ͱ‫ʹإ‬ΑΔ. ΅͕͢ɼ‫ࡏݱ‬ͷ൜ࡑ૞ࠪͰ͸‫إ‬΍෰૷ΛࡱӨ͢Δ͜ͱ͕ॏ. ϚϧνϞʔμϧ‫ݸ‬ਓೝূ [13–17] ͕༗ޮͳख๏ͱͯ͠ఏҊ. ཁͰ͋ΔͨΊɼ࣌ؒղ૾౓Λ٘ਜ਼ʹͯ͠Ͱ΋ߴۭؒղ૾౓. 1 a) b) c) d). େࡕେֶ Osaka University [email protected] [email protected] muramatsu@@am.sanken.osaka-u.ac.jp [email protected]. ⓒ 2014 Information Processing Society of Japan. Ͱ͋Δ͜ͱ͕ॏཁͱ͞Ε͍ͯΔ)ɽ‫إ‬΍਎௕͸੩తͳಛ௃Ͱ ͋ΔͨΊ࣌ؒղ૾౓ʹ͸͋·ΓӨ‫ࢥͱ͍ͳ͠ڹ‬ΘΕΔ͕ɼ า༰͸ 1 प‫ظ‬Λಛ௃ͱͯ͠༻͍ΔͨΊେ͖͘Ө‫͢ڹ‬Δɽ ͦ͜Ͱɼຊ࿦จͰ͸ɼา༰ɾ‫إ‬ɾ਎௕ʹΑΔϚϧνϞʔ. 1.

(2) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. (a) 30, 640×480 (b) 30, 80×60 (c) 3, 640×480 (d) 3, 80×60 (a) ‫ݪ‬ը૾ (b) γϧΤοτը૾ ਤ 1 าߦঢ়‫گ‬ͷ‫ݪ‬ը૾ (a) ͱɼϨϯζͷ࿪ΈΛऔΓআ͍ͨ‫ޙ‬ɼนʹ ਖ਼ର͢ΔΑ͏ʹม‫׵‬ΛߦͬͨγϧΤοτը૾ (b)ɽγϧΤοτ. ਤ 2. GEI ͷྫ (࣌ؒղ૾౓ [fps] ͱۭؒղ૾౓ [ըૉ] ͷ૊ΛԼʹ ࣔ͢).. ͷαΠζ͸ඃ‫ʹऀݧ‬ΑΔ͕ɼ90×180 ըૉఔ౓Ͱ͋Δ.. μϧ‫ݸ‬ਓೝূΛɼۭؒղ૾౓΍࣌ؒղ૾౓ͱ͍ͬͨࡱӨ৚ ݅Λߟྀͨ͠είΞϨϕϧ౷߹ͷ࿮૊ΈͰ࣮‫͢ݱ‬ΔɽΑΓ ۩ମతʹ͸ɼೖྗө૾ͷۭ࣌ؒղ૾౓ʹԠͯ͡ɼา༰ɾ‫إ‬ɾ ਎௕ʹΑΔείΞ΁ͷॏΈ෇͚Λద੾తʹมԽͤ͞Δํ๏ ΛఏҊ͢Δɽ୅දతͳۭ࣌ؒղ૾౓Λ‫ؚ‬ΉֶशσʔλΛ༻ ͍ͯ࠷దͳॏΈΛࢉग़͓͖ͯ͠ɼͦΕҎ֎ͷۭ࣌ؒղ૾౓ ͷςετσʔλʹରͯ͠͸ɼֶशσʔλ͔Βࢉग़ͨ͠ॏΈ Λ಺ૠ͢Δ͜ͱʹΑͬͯॏΈΛਪఆ͢Δɽ ͜ͷΑ͏ͳ໨తͷԼɼຊ࿦จͷߩ‫ݙ‬͸ɼ࣍ͷࡾͭͷ఺ʹ ·ͱΊΒΕΔɽ. 1. ༷ʑͳۭ࣌ؒղ૾౓ʹ͓͚ΔϚϧνϞʔμϧείΞσʔ ληοτͷߏங େ‫ن‬໛า༰σʔλϕʔε The OU-ISIR Gait Database,. the large population data set [20] ͷ 1,935 ਓͷσʔλΛ༻ ͍ͯา༰ɾ‫إ‬ɾ਎௕ͷείΞσʔληοτΛߏங͢Δɽ͜. ‫ݪ‬ը૾:640×480 ‫ݪ‬ը૾:320×240 ‫ݪ‬ը૾:160×120 ‫ݪ‬ը૾:80×60 ‫إ‬ը૾:24×22 ‫إ‬ը૾:12×10 ‫إ‬ը૾:6×6 ‫إ‬ը૾:3×2 ਤ 3 ༷ʑͳۭؒղ૾౓ͷ‫ݪ‬ը૾͔Β࡞੒ͨ͠‫إ‬ͷը૾ɽ‫إ‬ը૾ͷα Πζ͸ඃ‫ʹऀݧ‬ґଘ͢Δ͕ɼ͜͜Ͱͷը૾ͷαΠζ [ըૉ] Λ ͍ࣔͯ͠Δɽ. ͔Β 94 ࡀ·Ͱͷ෯޿͍೥୅૚ͷஉঁ߹‫ ܭ‬4,016 ਓͷาߦ ө૾͔ΒͳΔɽө૾ͷۭؒղ૾౓͸ 640×480 ըૉͰɼ࣌ ؒղ૾౓͸ 30fps Ͱ͋Δɽ‫إ‬ͷը૾͸ਤ 1(a) ͷը૾ྻ͔Β ࡞੒͢ΔɽγϧΤοτը૾͸ɼϨϯζͷ࿪ΈΛऔΓআ͍ͨ ‫ޙ‬ɼนʹਖ਼ର͢ΔΑ͏ʹม‫׵‬Λߦ͍ɼഎ‫ࠩܠ‬෼ʹ‫ͮ͘ج‬ά ϥϑΧοτηάϝϯςʔγϣϯʹΑͬͯ࡞੒ͨ͠ɽ͜ͷγ ϧΤοτ͸า༰ͷಛ௃நग़΍਎௕ͷ‫͍༻ʹࢉܭ‬Δɽ. ͜Ͱ͸༷ʑͳۭ࣌ؒղ૾౓ͷ૊Έ߹Θͤʹରͯ͠ɼείΞ. 2.2 είΞͷࢉग़. Λ‫͢ࢉܭ‬Δɽ. 2.2.1 า༰. 2. า༰ɾ‫إ‬ɾ਎௕ͷೝূਫ਼౓ͷۭ࣌ؒղ૾౓ʹର͢Δ‫౓ײ‬ ධՁ. ຊ‫Ͱڀݚ‬͸ɼา༰ಛ௃ͱͯ͠ɼา༰ೝূʹ͓͍ͯ෯޿͘ ར༻͞Ε͍ͯΔ GEI [21] Λ༻͍Δɽ·ͣɼਤ 1 ͷγϧΤο. ߏஙͨ͠σʔληοτΛ༻͍ͯา༰ɾ‫إ‬ɾ਎௕ͷͦΕͧ. τը૾͔Β 88×128 ըૉʹਖ਼‫ن‬Խͨ͠า༰γϧΤοτϘ. Εͷೝূਫ਼౓ͷධՁΛߦ͏ɽ۩ମతʹ͸ɼ1 ର 1 ೝূ໰୊. ϦϡʔϜ (Gait Silhouette Volume, GSV) Λ࡞੒͠ɼGSV. ʹ͓͚Δ౳Ձ‫ޡ‬Γ཰ (Equal Error Rate, EER) ͕ۭ࣌ؒղ. Λ 1 प‫Ͱظ‬ฏ‫ۉ‬ԽΛ͢Δ͜ͱʹΑΓਤ 2 ͷΑ͏ͳ GEI Λ. ૾౓ʹԠͯ͡ͲͷΑ͏ʹมԽ͢Δ͔Λղੳ͢Δɽ. ࡞੒͢Δɽϓϩʔϒ (ೖྗσʔλ) ͱΪϟϥϦ (ొ࿥σʔλ). 3. ۭ࣌ؒղ૾౓ʹదԠతͳείΞ౷߹ ·ͣɼۭ࣌ؒղ૾౓ʹదԠతͳॏΈ෇͚Λ͢Δʹ͋ͨͬ ͯɼֶशσʔλͷείΞσʔληοτʹରͯ͠୅දతͳε ίΞϨϕϧ౷߹ख๏Ͱ͋Δઢ‫ܗ‬ϩδεςΟοΫճ‫( ؼ‬Linear. Logistic Regression, LLR) Λ༻͍ͯ࠷దͳॏΈ෇͚Λ‫ࢉܭ‬. ͷ GEI ΛͦΕͧΕ Gp ɼGg ͱ͢ΔͱɼϓϩʔϒɼΪϟϥ Ϧͷ૬ҧ౓είΞ͸ϢʔΫϦου‫࣍ͯ͠ͱ཭ڑ‬ͷΑ͏ʹ‫ܭ‬ ࢉ͞ΕΔɽ. Sgait = ||Gp − Gg ||2 ,. ͢ΔɽߋʹɼֶशσʔλҎ֎ͷۭ࣌ؒղ૾౓ʹରͯ͠͸ɼ. ͜͜Ͱɼ|| · ||2 ͸ L2 ϊϧϜΛද͢ɽ. ֶशσʔλʹର͢ΔॏΈΛ಺ૠʹΑΓɼॏΈΛਪఆ͢Δख. 2.2.2 ‫إ‬. ๏ΛఆࣜԽ͢Δɽ·ͨɼͦͷఆࣜԽʹ‫͍ͯͮج‬ɼ༷ʑͳ࣌ ۭؒղ૾౓ʹର͢ΔςετσʔλͰਫ਼౓ධՁΛߦ͏ɽ. (1). ‫إ‬ೝূΛߦ͏ࡍɼ൅΍ྠֲΛऔΓআ͍ͨ‫إ‬ͷ෦෼Λ༻͍ ͯߦ͏͜ͱ͕ҰൠతͰ͋Δ͕ɼۭؒղ૾౓͕‫ʹ୺ۃ‬௿͍৔ ߹ (25×25 ըૉΑΓখ͍͞৔߹) ʹ͸ߴ͍ೝূਫ਼౓Λอͭ. 2. ༷ʑͳۭ࣌ؒղ૾౓ʹ͓͚ΔϚϧνϞʔμ ϧείΞσʔληοτ. ͜ͱ͕Ͱ͖ͳ͍ͨΊɼ൅΍ྠֲ΋‫ؚ‬Ίͨશମతͳಛ௃Λར. 2.1 าߦө૾σʔλϕʔε. ʹɼ಄ͷྖҬͷΧϥʔը૾͸γϧΤοτͷϚεΫΛՃ͑ͨ. ༻ͨ͠ํ͕ྑ͍ͱߟ͑ΒΕΔɽ·ͨɼਤ 3 ͔Β෼͔ΔΑ͏. ༷ʑͳۭ࣌ؒղ૾౓ʹ͓͚ΔϚϧνϞʔμϧσʔληο. ΋ͷͱͳ͍ͬͯΔɽ͜ͷ಄ͷྖҬͷը૾ͷαΠζ͸ඃ‫ऀݧ‬. τΛ࡞੒͢ΔͨΊɼThe OU-ISIR Gait Database, the large. ʹΑͬͯҟͳΔ͕ɼۭؒղ૾౓ΛԼ͛Δલͷ 640×480 ը. population data set Λ༻͍ͨɽ͜ͷσʔλϕʔεͰ͸ 1 ࡀ. ૉͷը૾Ͱ͸ 18×20 ըૉ͔Β 31×25 ըૉͷൣғͰ͋Δɽ. ⓒ 2014 Information Processing Society of Japan. 2.

(3) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report ද 1. ຊ࿦จͰ͸͜ͷ಄ͷྖҬΛ‫Ϳݺͱإ‬΋ͷͱ͢Δɽೋͭͷ‫إ‬. ۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓ [fps] ͷ૊.. ը૾ͷ૬ҧ౓͸ςϯϓϨʔτϚονϯάΛ༻͍ͯਖ਼‫ن‬Խ૬. SR. TR. ‫ޓ‬૬ؔͰ࣍ͷΑ͏ʹ‫ࢉܭ‬Λߦ͏ɽFpi Λ ϓϩʔϒ ͷ i ൪. 640 × 480, 480 × 360ɼ320 × 240, 213 × 160ɼ. 30, 15, 10ɼ. 160 × 120ɼ128 × 96ɼ106 × 80, 91 × 68ɼ. 7.5ɼ6ɼ5ɼ. 80 × 60ɼ64 ×48ɼ53 × 40ɼ40 × 30ɼ20 × 15ɼ. 3.75ɼ3ɼ2ɼ1. ໨ͷϑϨʔϜͷಛ௃ͱ͠ɼFgi,k Λ ΪϟϥϦͷ j ൪໨ͷϑ ϨʔϜͷɼςϯϓϨʔτϚονϯάͷ୳ࡧൣғͰͷ k ൪໨ ͷಛ௃ͱ͢Δɽͦͷͱ͖ɼϓϩʔϒͱ ΪϟϥϦͷ૬ҧ౓ε ίΞ͸࣍ͷΑ͏ʹ‫͞ࢉܭ‬ΕΔɽ. Sf ace = min[1 − N CC(Fpi , Fgj,k )], i,j,k. (2). ͜͜ͰɼN CC(Fpi , Fgj,k ) ͸ Fpi ͱ Fgj,k ͷؒͷਖ਼‫ن‬Խ૬‫ޓ‬ ૬ؔͰ͋Δɽ. 2.2.3 ਎௕ ਎௕͸ɼγϧΤοτͷ಄௖఺ͱ଍Լ఺ʹ‫͢ࢉܭ͍ͯͮج‬ Δɽ2.1 અͰड़΂ͨΑ͏ʹɼγϧΤοτը૾͸น໘ʹਖ਼ର ͢ΔΑ͏ʹม‫͞׵‬Ε͓ͯΓɼ·ͨɼาߦऀ͕น໘ͱฏߦͳ ҰఆͷίʔεΛา͍͍ͯΔ͜ͱ͔Βɼน໘ͱਖ਼ର͢ΔΧϝ ϥ͔Β‫ͨݟ‬Ԟߦ͖͸ɼҰఆͰ͋Δͱߟ͑Δ͜ͱ͕Ͱ͖Δɽ ΑͬͯɼγϧΤοτྖҬͷ֎઀ۣ‫ܗ‬ͷߴ͞ [ըૉ] ͔ΒɼΧ ϝϥߍਖ਼ʹΑͬͯಘΒΕΔΧϝϥύϥϝλͱԞߦ͖৘ใʹ ‫͍ͯͮج‬ɼੈք࠲ඪʹ͓͚Δ਎௕ [m] ʹม‫͢׵‬Δ͜ͱ͕Մ ೳͰ͋Δɽ·ͨɼ਎௕͸੩తͳಛ௃Ͱ͋ΔͨΊϑϨʔϜຖ ʹ‫͍ͯ͠ࢉܭ‬Δ͕ɼาߦಈ࡞ʹΑΔ্ԼಈʹΑΓଟগͷม Խ͕ଘࡏ͢ΔɽͦͷͨΊɼ਎௕ h ͸ҎԼͷΑ͏ʹը૾ྻʹ ରͯ͠ฏ‫ۉ‬ΛऔΔ͜ͱʹΑΓ‫͍ͯ͠ࢉܭ‬Δɽ Nf 1 h= Zi , Nf. (b) ‫إ‬ ਤ 4. (c) ਎௕. ֤ϞμϦςΟͷ ROC ‫ۂ‬ઢ. ࣍ʹɼ࣌ؒղ૾౓ʹ͍ͭͯઆ໌͢Δɽຊ࿦จ͸ 30fps Λ ࠷େͱͨ͠ 10 ௨ΓΛ༻͍ͨɽ30fps ͷը૾ྻ͔Β࣌ؒղ૾ ౓ʹԠͯ͡ҰఆͷִؒͰը૾ΛؒҾ͘͜ͱʹΑΓɼ௿࣌ؒ ղ૾౓ͷը૾ྻΛ࡞੒͢Δɽ ·ͨɼ࣌ؒղ૾౓Λμ΢ϯαϯϓϧͨ͠৔߹ɼ։࢝ϑϨʔ ϜʹΑͬͯը૾ྻ͕มԽ͢Δɽྫ͑͹ 15fps ͷ৔߹Λߟ͑ Δͱɼ‫਺ح‬൪໨ͷը૾Λ࢖༻͢Δ͔ɼ‫਺ۮ‬൪໨ͷը૾Λ࢖ ༻͢Δ͔Ͱ݁Ռ͕ҟͳΔɽͦͷͨΊɼ15fpsɼ10fpsɼ7.5fpsɼ. 6fpsɼ5fpsɼ3.75fpsɼ3fpsɼ2fpsɼ1fps ʹରͯ͠ɼͦΕͧΕ 2ɼ3ɼ4ɼ5ɼ6ɼ8ɼ10ɼ15ɼ30 छྨͷ։࢝ϑϨʔϜʹରԠ ͢Δ‫ݸ‬ผͷը૾ྻΛ࡞੒ͯ͠࢖༻͢ΔɽҎԼͰ͸͜ͷ։࢝. (3). i=1. ͜͜ͰɼZi ͸ i ൪໨ͷϑϨʔϜͷ਎௕ɼNf ͸ը૾ྻͷϑ ϨʔϜ਺Ͱ͋Δɽ. hp ɼhg ΛͦΕͧΕϓϩʔϒɼΪϟϥϦͷ਎௕ͱ͢Δͱɼ ૬ҧ౓είΞ Sheight ͸ઈର஋Λ༻͍ͯҎԼͷΑ͏ʹ‫ࢉܭ‬ ͢Δɽ. Sheight = |hp − hg |,. (a) า༰. ϑϨʔϜͷछྨΛ NT R ͱͯ͠ද͢ɽ. 2.2.5 είΞσʔληοτ ຊ‫ߏͰڀݚ‬ங͢ΔείΞσʔληοτʹ͸ɼେ‫ن‬໛า༰ σʔλϕʔε͔Β 1,935 ਓͷαϒηοτΛબ୒͠ɼͦΕͧ Εͷඃ‫ ͍ͯͭʹऀݧ‬85 ౓ͷ‫؍‬ଌํ޲ (΄΅ଆ໘ํ޲) ͔Β ࡱӨ͞ΕͨɼϓϩʔϒͱΪϟϥϦ ͷ૊Έ߹ΘͤΛ࢖༻͠ ͨɽΪϟϥϦʹؔͯ͠͸ 1,935 ௨Γɼϓϩʔϒʹؔͯ͠͸. (4). 2.2.4 ۭ࣌ؒղ૾౓ͷμ΢ϯαϯϓϦϯά. 1,935NT R ௨Γͷσʔλʹ͍ͭͯɼۭؒղ૾౓͕ 13 ௨Γɼ ࣌ؒղ૾౓͕ 10 ௨Γɼͭ·Γ߹Θͤͯ 130 ௨ΓͷείΞΛ ‫͢ࢉܭ‬Δɽ͜ͷείΞσʔλ͸ɼΪϟϥϦɼϓϩʔϒͷඃ‫ݧ‬. ‫ݪ‬ը૾ͱγϧΤοτը૾Λۭ࣌ؒղ૾౓ʹؔͯ͠μ΢ϯ. ऀ ID Ϧετͷ͢΂ͯͷ૊Έ߹Θͤʹؔ͢Δ૬ҧ౓είΞߦ. αϯϓϧ্ͨ͠Ͱɼา༰ɾ‫إ‬ɾ਎௕ͦΕͧΕʹ͍ͭͯεί. ྻͷ‫Ͱܗ‬ද͞ΕΔɽͭ·Γɼຊਓಉ࢜ͷείΞ͸ 1,935NT R. ΞσʔληοτΛ࡞੒͢Δɽۭ࣌ؒղ૾౓ͷόϦΤʔγϣ. ௨ΓɼଞਓͱͷείΞ͸ 1,935NT R ×1,934=3,742,290NT R. ϯ͸ɼද 1 ʹࣔ͢௨ΓͰ͋Δɽ·ͨɼຊ࿦จͰ͸ɼදதͳ. ௨Γଘࡏ͢Δɽ͜ͷߦྻ͸า༰ɾ‫إ‬ɾ਎௕ͦΕͧΕʹ͍ͭ. ͲͰɼۭؒղ૾౓Λ SRɼ࣌ؒղ૾౓Λ TR ͱུ‫͢ه‬Δɽ. ͯɼۭ࣌ؒղ૾౓ͷ͢΂ͯͷ૊Έ߹Θͤʹ͍ͭͯ‫͢ࢉܭ‬Δɽ. ·ͣɼۭؒղ૾౓ʹ͍ͭͯઆ໌͢Δɽຊ࿦จͰ͸‫ݪ‬ը૾ ͷ 680×480 ըૉΛ࠷େͱͨ͠߹‫ ܭ‬13 ௨ΓΛ༻͍ͨɽ͜Ε. 3. า༰ɾ‫إ‬ɾ਎௕ͷೝূੑೳධՁ. ʹΑΓɼ༷ʑͳਓ෺αΠζͷา༰ɾ‫إ‬ɾ਎௕ಛ௃͕ಘΒΕΔ. 3.1 ̍ର̍ೝূʹର͢Δ݁Ռ. ͨΊɼΧϝϥ‫ݻ‬༗ͷը૾αΠζͷҧ͍ʹՃ͑ͯɼΧϝϥ͔ Βਓ෺·Ͱͷ‫ʹ཭ڑ‬Ԡͨ͡ਓ෺αΠζͷมԽʹ͍ͭͯ΋ߟ ྀʹೖΕͨղੳΛߟ͑Δ͜ͱ͕Ͱ͖Δɽۭؒղ૾౓Λμ΢ ϯαϯϓϧͨ͠ GEI ͱ‫إ‬ͷը૾ͷྫΛਤ 2 ͱਤ 3 ʹࣔ͢ɽ. ⓒ 2014 Information Processing Society of Japan. ͜͜Ͱ͸ 1 ର 1 ೝূʹ͓͚Δา༰ɾ‫إ‬ɾ਎௕ͷ‫ݸ‬ผͷೝ ূੑೳͷ݁ՌΛࣔ͢ɽ ධՁࢦඪͱͯ͠ɼଞਓडೖ‫ޡ‬Γ཰ (False Acceptance Rate,. FAR) ͱຊਓ‫ڋ‬൱‫ޡ‬Γ཰ (False Rejection Rate, FRR) ͷτ. 3.

(4) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report ද2. ද3. ͢΂ͯͷۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓ [fps] ͷ૊Έ߹Θͤʹ. ͓͚Δ EER [%] (า༰)ɽ SR\TR 30 15 10 7.5. ͢΂ͯͷۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓ [fps] ͷ૊Έ߹Θͤʹ ͓͚Δ EER [%] (‫)إ‬ɽ”-”͸νϟϯεϨϕϧΛද͠ɼ50%Ͱ. 6. 5. 3.75. 3. 2. 1. 640×480. 2.4. 2.5. 2.6. 3.7. 7.4. 9.6. 22.0 23.9 39.4 39.5. ͋Δ͜ͱΛද͢. SR\TR 30 15. 10. 7.5. 6. 480×360. 2.4. 2.5. 2.6. 3.7. 7.5. 9.8. 22.1 24.1 39.6 39.4. 640×480. 4.5. 5.8. 7.2. 8.0. 7.7. 320×240. 2.4. 2.4. 2.7. 3.9. 7.5. 9.9. 22.1 24.3 39.7 39.4. 480×360. 3.1. 3.8. 5.2. 5.9. 6.2. 8.0. 213×160. 2.3. 2.5. 2.8. 4.2. 8.0. 10.4 22.2 25.0 39.8 39.8. 320×240. 4.6. 5.8. 7.7. 8.7. 8.7. 10.7 10.8 10.8 10.8 11.1. 160×120. 2.5. 2.6. 3.1. 4.4. 8.3. 10.7 22.6 25.9 40.0 39.6. 213×160. 3.2. 3.9. 4.8. 6.4. 7.4. 9.7. 128×96. 2.9. 3.2. 3.7. 4.9. 8.9. 11.4 23.2 26.7 40.4 40.1. 160×120. 4.6. 5.4. 6.6. 8.4. 9.8. 12.2 12.6 12.8 12.5 13.7. 106×80. 2.6. 2.9. 3.9. 5.9. 9.7. 12.0 23.4 27.4 39.8 39.6. 128×96. 7.2. 8.1. 9.5. 11.7 12.9 15.3 15.8 16.2 16.2 17.6. 91×68. 2.2. 2.6. 5.1. 6.5. 9.9. 12.9 23.7 28.2 39.9 39.9. 106×80. 10.0 11.1 13.1 16.0 17.8 20.3 20.4 21.1 21.0 23.0. 80×60. 2.6. 4.2. 6.9. 6.3. 11.3 14.1 23.3 28.8 40.1 39.4. 91×68. 13.6 15.0 16.9 20.3 21.8 25.1 25.1 25.6 25.3 27.6. 64×48. 2.9. 11.2. 5.0. 12.5 12.0 16.4 24.3 29.9 39.8 39.0. 80×60. 22.7 23.6 25.7 27.3 29.1 30.2 31.0 30.8 31.6 32.9. 53×40. 3.8. 8.0. 9.0. 10.9 19.4 17.8 26.0 31.8 40.8 40.1. 64×48. 23.3 24.7 27.1 28.4 32.3 32.8 33.3 34.2 33.9 34.4 38.3 36.3 36.8 37.0 38.2 39.5 39.8 40.0 39.7 40.1. 5. 3.75. 3. 2. 1. 10.0 10.1 10.3 10.2 10.5 8.0 9.8. 8.1 9.9. 8.1. 8.3. 9.7. 10.5. 40×30. 5.7. 23.7 17.3 18.5 23.9 28.9 31.3 42.0 38.9. 53×40. 20×15. 18.1 20.2 25.3 25.8 28.0 24.8 30.4 32.7 36.3 32.0. 40×30. -. -. -. -. -. -. -. -. -. -. 20×15. -. -. -. -. -. -. -. -. -. -. 7.7. ϨʔυΦϑΛࣔ͢ड৴ऀૢ࡞ಛੑ (Receiver Operatorating. Characteristic, ROC) ‫ۂ‬ઢΛ༻͍Δɽਤ 4 ʹ (1) ߴۭ࣌ؒ. ද4. ͢΂ͯͷۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓ [fps] ͷ૊Έ߹Θͤʹ. ղ૾౓ɼ(2) ߴۭؒղ૾౓ͱ௿ۭ࣌ؒղ૾౓ɼ(3) ௿ۭؒղ. ͓͚Δ EER [%] (਎௕)ɽ SR\TR 30 15 10 7.5. ૾౓ͱߴ࣌ؒղ૾౓ɼ(4) ௿ۭ࣌ؒղ૾౓ͷ 4 ௨Γʹର͢Δ. 640×480 16.2 16.5 17.0 17.8 17.6 19.4 19.4 19.4 19.4 19.4. ROC ‫ۂ‬ઢΛࣔ͢ɽ͞Βʹɼۭ࣌ؒղ૾౓ͷ͢΂ͯͷ૊Έ. 480×360 16.3 16.6 16.9 18.0 17.8 19.6 19.6 19.6 19.6 19.6. ߹Θͤʹ͓͚Δ FAR ͱ FRR ͷ౳Ձ‫ޡ‬Γ཰ EER Λද 2-4. 320×240 16.5 16.9 17.0 18.0 17.8 19.9 20.0 20.0 19.8 20.0. ʹࣔ͢ɽۭؒղ૾౓͕‫ʹ୺ۃ‬௿͍৔߹ɼ‫إ‬ը૾͕ 1×1 ըૉ. 6. 5. 3.75. 3. 2. 1. 213×160 16.3 16.8 17.4 18.3 18.4 20.0 20.1 20.1 20.1 20.1 160×120 16.5 17.2 17.6 18.6 18.2 20.9 21.0 21.0 21.0 21.0. ʹͳͬͯ͠·͏৔߹͕ଘࡏ͢ΔɽͦͷΑ͏ͳ৔߹ʹ͸ɼਖ਼. 128×96. 17.0 17.7 18.1 19.3 19.1 21.7 21.7 21.7 21.7 21.7. ‫ن‬Խ૬‫ޓ‬૬ؔͰείΞΛ‫ͨ͏·ͯͬ͠ͳ͘ͳ͖Ͱࢉܭ‬Ίɼ. 106×80. 17.3 18.2 18.8 20.1 20.3 23.1 23.2 23.2 23.3 23.2. νϟϯεϨϕϧͱͯ͠ѻ͍ɼEER Λ 50%ͱ͍ͯ͠Δɽ. 91×68. 15.8 16.8 17.6 19.8 19.9 22.9 23.0 23.0 23.0 23.0. ͜ͷ݁Ռ͔Β෼͔ΔΑ͏ʹɼ‫إ‬ೝূͷੑೳ͸ۭؒղ૾౓. 80×60. 18.2 19.4 21.7 23.2 23.3 27.3 27.3 27.3 27.1 27.3. ͕Լ͕ͬͨͱ͖ʹେ͖͘௿Լ͍ͯ͠Δ͜ͱ͕෼͔Δɽͦͷ. 64×48. 15.7 18.3 21.4 24.6 24.4 29.7 29.7 29.4 29.7 29.7. 53×40. 18.0 21.8 24.9 28.2 28.9 33.2 33.2 33.2 33.1 33.2. 40×30. 19.6 24.8 30.0 33.6 33.4 38.7 38.9 38.8 38.8 38.8. 20×15. 31.6 37.2 41.0 42.2 41.4 43.1 43.2 43.0 43.0 43.0. ҰํͰา༰ೝূͷੑೳ͸ۭؒղ૾౓͕Լ͕ͬͯ΋͋·Γམ ͪͣɼ࣌ؒղ૾౓͕Լ͕ͬͨͱ͖ʹେ͖͘௿Լ͍ͯ͠Δ͜ ͱ͕෼͔Δɽ·ͨɼ਎௕ʹΑΔೝূ͸ɼಉ͡਎௕ͷਓ͕ଟ ͘ଘࡏ͢Δ͜ͱ͔Βา༰΍‫ͱإ‬ൺֱ͢Δͱશମతʹੑೳ͸ ௿͍ɽͦͯ͠ɼۭ࣌ؒղ૾౓ͷͲͪΒ͔͕௿͘ͳͬͯ΋͋ ·Γੑೳ͕མͪͣɼ྆ํ͕௿͘ͳͬͨͱ͖ʹੑೳ͕௿Լ͠ ͍ͯΔ͜ͱ͕෼͔Δɽ. 3.2 ‫ݸ‬ʑͷϞμϦςΟʹର͢Δߟ࡯ ͦΕͧΕͷಛ௃ͷ܏޲Λ෼ੳ͢ΔͨΊʹ࣌ؒղ૾౓ɼۭ ؒղ૾౓ΛͦΕͧΕ‫ݻ‬ఆͨ͠৔߹ͷ EER ͷมԽΛਤ 5 ͱ ਤ 6 ʹࣔ͢ɽ ߴ࣌ؒղ૾౓ (30fps, ਤ 5) ʹ͓͚Δۭؒղ૾౓ʹର͢ Δ EER ͷมԽΛ‫ݟ‬Δͱɼ160×120 ըૉΑΓۭؒղ૾౓͕. (a) 30fps ਤ 5. (b) 3fps. ࣌ؒղ૾౓Λ‫ݻ‬ఆͨ͠ͱ͖ͷۭؒղ૾౓ʹΑΔ EER ͷมԽ. Լ͕ͬͯ͘Δͱɼ‫إ‬ೝূͷੑೳ͕େ͖͘ѱԽ͍ͯ͠Δ͜ ͱ͕෼͔Δɽਤ 3 (c) Λ‫ͯݟ‬΋෼͔ΔΑ͏ʹɼ‫إ‬ͷը૾͸. ੑೳ͕ѱԽ͍ͯ͠Δɽ. 160×120 ըૉ͔ΒαΠζ͕͔ͳΓখ͘͞ͳ͍ͬͯΔɽͦ. ௿࣌ؒղ૾౓ (3fps) ʹ͓͚Δۭؒղ૾౓ʹର͢Δ EER. ͷҰํͰɼา༰ͱ਎௕͸ɼۭؒղ૾౓ͷதఔ౓·Ͱͷ௿Լ. ͷมԽʹ͓͍ͯ΋ɼߴ࣌ؒղ૾౓ͷͱ͖ͱಉ͡Α͏ͳ܏޲. ʹରͯ͋͠·Γେ͖͘͸ੑೳ͕ѱԽ͍ͯ͠ͳ͍͜ͱ͕෼͔. ͕‫ݟ‬ΒΕΔɽۭؒղ૾౓͕Լ͕ΔʹͭΕͯ‫إ‬ೝূͷੑೳ͕. Δɽ53×40 ըૉΛԼճΔΑ͏ͳղ૾౓͕‫ۃ‬Ίͯ௿͍৔߹ʹ. େ͖͘ѱԽ͍ͯ͠Δ͕ɼߴ࣌ؒղ૾౓ (30fps) ͷ࣌ΑΓ΋. ͸ɼਓ෺αΠζ͕໿ 8×15 ըૉ΄Ͳʹͳͬͯ͠·͏ͨΊɼ. ߴ͍ۭؒղ૾౓͔ΒੑೳѱԽ͕‫ݟ‬ΒΕΔɽ͜Ε͸ߴ࣌ؒղ. ⓒ 2014 Information Processing Society of Japan. 4.

(5) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report ද 5. ֶशσʔλͱςετσʔλͷۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓. [fps] ͷ૊. σʔληοτ SR Training Test. TR. 640 × 480, 320 × 240, 160 × 120, 30, 15, 7.5, 106 × 80, 80 × 60, 53 × 40. 5, 3, 1. 480 × 360, 213 × 160,. 10, 6,. 128 × 96, 91 × 68, 64 ×48. 3.75, 2. ·ͨɼ͜ͷΑ͏ͳॏΈ͸͍͔ͭ͘ͷ୅දతͳۭ࣌ؒղ૾ ౓ͷ૊ (ֶशσʔλ) ʹରͯ͠‫ٻ‬Ί͓͖ͯɼͦΕҎ֎ͷۭ࣌ ؒղ૾౓ͷ૊ (ςετσʔλ) ʹରͯ͠͸಺ૠʹΑΔਪఆΛ (a) 640×480 ըૉ (b) 80×60 ըૉ ਤ 6 ۭؒղ૾౓Λ‫ݻ‬ఆͨ͠ͱ͖ͷ࣌ؒղ૾౓ʹΑΔ EER ͷมԽ. ૾౓ͷ৔߹͸ɼը૾ͷαϯϓϦϯάຕ਺͕ଟ͍͜ͱ͔Βۭ ؒղ૾౓ͷ௿Լʹରͯ͠ଟগͷੑೳѱԽͷ཈੍ʹͭͳ͕Δ ΋ͷͷɼ3fps ͷΑ͏ͳ௿࣌ؒղ૾౓ͷͱ͖ʹ͸ɼαϯϓϦ ϯάຕ਺ͷগͳ͔͞Βɼۭؒղ૾౓ͷ௿Լ͕ΑΓ௚઀తʹ ੑೳѱԽʹͭͳ͕ΔͨΊͱߟ͑ΒΕΔɽ ·ͨɼ਎௕ʹ͍ͭͯͷ݁ՌΛ‫ͯݟ‬ΈΔͱɼ3fps ͷͱ͖͸ ۭؒղ૾౓ʹΑΔӨ‫ڹ‬Λେ͖͘ड͚͍ͯΔ͕ɼ30fps ͷͱ͖ ͸ۭؒղ૾౓͕Լ͕ͬͯ΋͋·Γੑೳ͕ѱԽ͍ͯ͠ͳ͍͜ ͱ͕෼͔Δɽ͜ͷ͜ͱ͸ɼ਎௕͕ۭؒղ૾౓ʹΑͬͯେ͖ ͳӨ‫ڹ‬Λड͚Δͱ͍͏ࣄલͷ༧૝ʹ൓͍ͯ͠Δ͕ɼ਎௕͸ ฏ‫ۉ‬Λͱͬͯ‫ࢉܭ‬Λߦ͍ͬͯΔͨΊɼาߦͷ্Լಈʹ൐͏ ߴ͞ͷมಈʹΑΓ͋Δछͷ௒ղ૾ͷΑ͏ͳޮՌ͕ಘΒΕɼ. ߦ͏ɽͦͷΑ͏ͳֶशσʔλͱςετσʔλͷۭ࣌ؒղ૾ ౓ͷ૊͸ද 5 ʹࣔͨ͠௨ΓͰ͋Δɽ͜͜Ͱɼ3.1 અͰड़΂ ͨΑ͏ʹɼ‫إ‬ೝূʹ͓͍ۭͯؒղ૾౓͕ 40×30 ըૉҎԼʹ ͳΔͱɼνϟϯεϨϕϧͱͳΔ͜ͱ͔ΒɼείΞ౷߹ͷ࣮ ‫͍͓ͯʹݧ‬͸ɼۭؒղ૾౓͸ 40×30 ըૉɼ20×15 ըૉΛ আ͍ͨ 11 ௨Γͱͳ͍ͬͯΔ఺ʹ஫ҙ͞Ε͍ͨɽ. 4.2 είΞϨϕϧ౷߹ͷ࿮૊Έ ·ͣɼείΞϨϕϧ౷߹Λߦ͏લʹͦΕͧΕͷϞμϦςΟ ͷείΞͷਖ਼‫ن‬ԽΛߦ͏ɽsm (i, j) Λ i ൪໨ͷΪϟϥϦɼj ൪ ໨ͷϓϩʔϒʹ͓͚ΔɼϞμϦςΟ m ∈ {f ace, gait, height} ͷείΞͱ͢Δɽͦͷͱ͖ਖ਼‫ن‬ԽείΞ s¯m (i, j) ͸࣍ͷΑ ͏ʹ‫͢ࢉܭ‬Δɽ. s¯m (i, j) =. ݁Ռͱͯ͠αϒϐΫηϧΦʔμʔͷ਎௕͕ಘΒΕ͍ͯΔ͜ ͱ͕‫ݪ‬ҼͱࢥΘΕΔɽ͜ΕʹΑΓɼߴ࣌ؒղ૾౓ͷ৔߹͸ ۭؒղ૾౓͕௿ͯ͘΋ੑೳΛߴ͘อͭ͜ͱ͕Ͱ͖͍ͯΔͱ ࢥΘΕΔɽ ࣍ʹɼߴۭؒղ૾౓ (640×480 ըૉ) ʹ͓͚Δ࣌ؒղ૾ ౓ʹΑΔ EER ͷมԽ (ਤ 6) Λ‫ݟ‬Δͱɼ‫ͱإ‬਎௕͸࣌ؒղ. sm (i, j) − μm (i) , σm (i). (5). ͜͜Ͱɼμm (i) ͱ σm (i) ͸ฏ‫ͱۉ‬ඪ४ภࠩͰ͋Δɽ ͦͯ͠ɼೋͭͷը૾ྻ͕༩͑ΒΕͨͱ͖ʹͦΕ͕ຊ ਓ ಉ ࢜ Ͱ ͋ Δ ࣄ ‫ ཰ ֬ ޙ‬Λ ɼͦ Ε ͧ Ε ͷ ਖ਼ ‫ ن‬Խ ε ί Ξ. s¯ = [¯ sf ace , s¯gait , s¯height ]T Λ༻͍ͯ‫͢ࢉܭ‬Δ (ຊਓಉ࢜ ͷࣄ৅Λ X = 1 ͱ͢Δ)ɽ͞ΒʹɼSR qS ͱ TR qT ͸ε. ૾౓ͷ௿ԼʹԠͯ͡؇΍͔ʹੑೳ͕ѱԽ͍ͯ͠Δͷʹର͠. ίΞʹӨ‫ڹ‬Λ༩͑ΔͨΊɼq = [qS , qT ]T ΛߟྀʹೖΕΔ. ͯɼา༰͸ 7.5fps ΛԼճΔͱେ͖͘ੑೳ͕ѱԽ͍ͯ͠Δɽ. ඞཁ͕͋Δɽͦ͜Ͱɼۭ࣌ؒղ૾౓ʹదԠతͳࣄ‫཰֬ޙ‬. ௿ۭؒղ૾౓ (80×60 ըૉ) ʹ͓͚Δ࣌ؒղ૾౓ʹର͢ Δ EER ͷมԽΛ‫ͯݟ‬΋ɼา༰ͱ਎௕ʹ͍ͭͯ͸ಉ͡Α͏ ͳ܏޲͕‫ݟ‬ΒΕΔɽ͔͠͠ɼ‫إ‬͸ߴۭؒղ૾౓ͷͱ͖ͱൺ ΂ͯɼ࣌ؒղ૾౓ʹΑΔӨ‫͕ڹ‬େ͖͍͜ͱ͕෼͔Δɽ. 4. ۭ࣌ؒղ૾౓ʹదԠతͳείΞ౷߹ 4.1 είΞ౷߹ͷ֓ཁ ຊઅͰ͸ɼߏஙͨ͠είΞσʔλϕʔε΍֤ϞμϦςΟ. P (X = 1|¯ s; q) Λ LLR [22] Λ༻͍ͯද‫͠ݱ‬ɼϩδοτؔ਺ Λ࣍ͷΑ͏ͳείΞͷॏΈ෇͚࿨Ͱද͢ɽ. log. P (X = 1|¯ s; q) 1 − P (X = 1|¯ s; q). =. αm (q)¯ sm+αc (q), (6). m∈{f ace, gait,height}. ͜͜Ͱ αm (q) ͸ϞμϦςΟ m ͷॏΈɼαc (q) ͸ఆ਺߲Ͱ ͋ΔɽॏΈͷ஋Ͱ͋Δ αf ace , αgait , αheight , αc ͸ֶशσʔ λΛ༻͍ͯɼۭ࣌ؒղ૾౓ͷ஋ʹԠͯ͡‫ࢉܭ‬Λߦ͏ɽ. ʹର͢Δੑೳղੳͷ݁Ռʹ‫͍ͯͮج‬ɼదԠతͳείΞ౷߹ ख๏Λಋೖ͢Δɽ·ͣɼ֤ۭ࣌ؒղ૾౓ʹର͢Δ࠷దͳॏ. 4.3 ॏΈͷਪఆ. ΈΛɼLLR ʹΑֶͬͯश͢Δɽৄࡉ͸ 4.2ɼ4.3 અʹࣔ͢. 1 અͰड़΂ͨΑ͏ʹɼ͢΂ͯͷۭ࣌ؒղ૾౓ͷ૊ʹ͍ͭͯ. ͕ɼLLR ͸ֶशσʔλΛඞཁͱ͢Δ͜ͱ͔Βɼ1,935 ਓͷ. ͋Β͔͡ΊॏΈΛ‫ͱ͓ͯ͘͜͠ࢉܭ‬͸‫࣮ݱ‬తͳํ๏Ͱ͸ͳ. ඃ‫ऀݧ‬ΛϥϯμϜʹֶशηοτͱςετηοτʹ෼ׂ͢. ͍ɽͦͷͨΊɼ୅දతͳۭ࣌ؒղ૾౓ʹର͢ΔॏΈΛ‫ࢉܭ‬. Δɼ2 ෼ׂަࠩ‫ݕ‬ఆΛ༻͍ͯධՁΛߦ͏ɽϥϯμϜͳ෼ׂ. ͓͖ͯ͠ɼςετσʔλͷۭ࣌ؒղ૾౓ q∗ ʹର͢ΔॏΈ α∗. ๏ʹΑͬͯਫ਼౓͕ҟͳΔͨΊɼ͜ͷධՁΛ 50 ճ‫܁‬Γฦ͠. Λਪఆ͢Δ͜ͱΛߟ͑Δɽ༗‫ݶ‬ͷֶशσʔλ D = [Q, α]T. ߦ͍ɼͦͷฏ‫ʹۉ‬ΑͬͯධՁΛߦ͏ɽ. Λ༻͍ɼN ‫ݸ‬ͷۭ࣌ؒղ૾౓ͷ૊ Q = {qi }(i = 1, . . . , N ). ⓒ 2014 Information Processing Society of Japan. 5.

(6) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ͱֶशσʔλΛ༻͍ͯ‫ͨ͠ࢉܭ‬ॏΈ α = [α1 , . . . , αN ]T ͕ ࣄલʹ༩͑ΒΕΔ΋ͷͱͯ͠ɼ֤ϞμϦςΟಠཱʹ͍ͭͯ. ද6. ֶशσʔλͷ͢΂ͯͷۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓ [fps] ʹ ͓͚Δ EER [%]ɽଠࣈ͸ SumɼLLR (Fixed)ɼLLR ͷதͰ ࠷΋ਫ਼౓͕ߴ͍΋ͷΛࣔ͢ɽ. ॏΈͷਪఆΛߦ͏ɽ ਤ 5ɼ6 ʹࣔͨ͠Α͏ʹɼͦΕͧΕͷϞμϦςΟͷਫ਼౓. SR. TR. Fusion rule Sum LLR (Fixed) LLR. ͸ඇઢ‫ʹܗ‬มԽ͢ΔͷͰɼॏΈͷ஋΋ඇઢ‫ͳܗ‬࿮૊Έʹͯ. 30. 0.8. 0.8. 0.8. ‫͢ࢉܭ‬Δɽ۩ମతʹ͸ɼඇઢ‫ܗ‬ͷΧʔωϧؔ਺ʹΑΔΨ΢. 15. 0.9. 0.9. 0.9. εաఔճ‫( ؼ‬GPR) Λ༻͍ɼֶशσʔλ D ͱۭ࣌ؒղ૾౓. 7.5. 1.5. 1.4. 1.5. 5. 3.6. 3.5. 3.3. 3. 9.6. 9.1. 6.0. 1. 16.6. 15.4. 7.6. 30. 0.8. 0.7. 0.7. 15. 0.8. 0.8. 0.8. 7.5. 1.5. 1.4. 1.5. 5. 3.9. 3.9. 3.6. 640×480. q∗ ͔ΒॏΈΛਪఆ͢Δɽ ·ͣɼೋͭͷۭ࣌ؒղ૾౓ qi ͱ qj ͷߴ࣍‫ݩ‬ಛ௃ۭؒͰ ͷ಺ੵΛද͢ɼϥδΞϧ‫ج‬ఈؔ਺ (radial basis function,. RBF)k ΛҎԼͷΑ͏ʹఆٛ͢Δɽ ||qi − qj ||2 k(qi , qj ; θ) = v exp − , 2r2. (7). 320×240. 3. 9.8. 9.4. 6.8. 1. 16.6. 15.8. ͦͷͱ͖ɼࣄ‫཰֬ޙ‬෼෍ P (α∗ | q∗ , D) ͸ҎԼͷΑ͏ͳฏ‫ۉ‬. 8.6. 30. μ∗ ɼඪ४ภࠩ σ∗2 ͷΨ΢ε෼෍ͱͳΔ [23]ɽ. 0.8. 0.8. 0.8. 15. 1.0. 1.0. 1.0. 7.5. 2.1. 2.3. 2.0. 5. 6.2. 6.9. 5.5. 3. 12.1. 12.7. 10.6. 1. 19.7. 19.8. 14.2. 30. 1.3. 1.4. 1.3. 15. 1.6. 1.8. 1.5. 7.5. 4.2. 4.8. 3.3. 5. 11.8. 13.3. 8.3. 3. 18.2. 19.4. 14.9. 1. 26.4. 27.2. 20.0. 30. 2.5. 2.9. 1.8. 15. 3.8. 4.2. 2.7. 7.5. 7.4. 8.2. 5.3. 5. 16.2. 17.7. Λද 6ɼROC ‫ۂ‬ઢΛਤ 10 ʹࣔ͢ɽ͜͜Ͱ͸ɼ୯७ͳεί. 12.1. 3. 23.0. 24.4. 19.7. ͜͜Ͱ θ = [v, r]T ͸Χʔωϧؔ਺ͷύϥϝʔλͰ͋Δɽ. μ∗ = k∗T (K + Σ)−1 α. (8). 2 σ∗2 = k(q∗ , q∗ ; θ) − k∗T (K + Σ)−1 k∗ + σo,∗ ,. (9). 160×120. ͜͜Ͱ K ͸ (i, j) ͷཁૉΛ k(qi , qj ; θ) ͱ͢Δ N × N ͷਖ਼ ํߦྻɼk∗ ͸ୈ i ߦΛ k(qi , q∗ ; θ) ྻϕΫτϧɼΣ ͸ (i, i) 2 ͷཁૉΛ σi2 ͱ͢Δ N × N ͷର֯ߦྻͰ͋Δɽσo,∗ ͸‫؍‬ଌ. ϊΠζͰ͋Δɽ. 106×80. ͜ΕΑΓɼ͋Δۭ࣌ؒղ૾౓ q∗ ʹର͠ɼฏ‫ ۉ‬μ∗ ΛॏΈ. α∗ ͱͯ͠༻͍Δɽ 4.4 ֶशσʔλͷධՁ ςετσʔλධՁʹઌཱͬͯɼֶशσʔλʹର͢Δ EER. 80×60. Ξ౷߹ख๏Ͱ͋Δ Sumɼۭ࣌ؒղ૾౓ʹؔ܎ͳ͘‫ݻ‬ఆ͠. 1. 31.2. 31.5. 25.5. ͨॏΈ෇͚Λ༻͍Δ LLR (Fixed) ͱɼۭ࣌ؒղ૾౓ͷ૊ʹ. 30. 5.1. 13.6. 3.3. ରͯ͠‫ݸ‬ผʹֶशͨ͠ LLR ͷ݁ՌΛൺֱ͢Δɽ͜͜Ͱ͸ɼ. 15. 9.8. 18.1. 7.0. ॏΈͷֶशͱੑೳධՁʹ༻͍Δۭ࣌ؒղ૾౓͕Ұக͍ͯ͠. 7.5. 15.2. 23.5. 10.2. 5. 21.3. 28.1. 16.3. 3. 26.7. 33.1. 25.5. 1. 32.2. 42.0. 31.0. Δ͜ͱ͔ΒɼLLR ͸ɼ࣍અͷςετσʔλͷධՁʹ͓͚Δ. LLR (GT) ʹ૬౰͢Δɽ. 53×40. LLR ͷੑೳ͸ɼSR ͱ TR ͷͲͪΒ΋ߴ͍ͱ͖΍௿͍ͱ ͖ (ਤ 10(a)ɼ(d)) ͸ Sum ͱ΄΅ಉ͡ਫ਼౓Ͱ͋Δ͕ɼͲͪ Β͔͕௿͍ͱ͖ (ਤ 10(b)ɼ(c)) ͸ Sum ΑΓྑ͘ͳ͍ͬͯ Δ͜ͱ͕෼͔Δɽ. LLR (Fixed) Ͱ͸‫إ‬ͷॏΈ͕า༰΍਎௕ͱൺ΂ͯେ͖ ͘ɼSR ͕௿͍ͱ͖͸‫إ‬ͷਫ਼౓͸͔ͳΓ௿͘ͳΔͨΊɼLLR. (Fixed) ͸ SR ͕௿͍ͱ͖͸࠷΋ਫ਼౓͕ѱ͘ͳ͍ͬͯΔ (ਤ 10(b)ɼ(d))ɽ ͞ΒʹɼEER ͷɼSR Λ‫ݻ‬ఆͨ͠ͱ͖ͷ TR ʹΑΔมԽ. 4.5 ςετσʔλͷධՁ ͜͜Ͱ͸ςετσʔλͷ 2 ෼ׂަࠩ‫ݕ‬ఆ͸ 5 ճ‫܁‬Γฦ͠ ߦͬͨɽ·ͨɼΨ΢εաఔճ‫͍͓ͯʹؼ‬ɼۭ࣌ؒղ૾౓͸ର ਺Ͱද͠ (ͨͱ͑͹൒෼ͷαΠζͰ͋Ε͹ qs = log(0.5))ɼ Χʔωϧؔ਺ͷύϥϝʔλ͸ r = 0.2ɼv = 1 ͱͨ͠ɽ ·ͨɼ2 ௨Γͷ࣌ؒղ૾౓ (10fps ͱ 2fps) ͱ 2 ௨Γͷۭ ؒղ૾౓ (3/4ɼ1/10) ͷ૊Έ߹Θͤ 4 ௨Γʹର͢Δ ROC ‫ۂ‬ઢΛਤ 10 ʹࣔ͢ɽ. ͱ TR Λ‫ݻ‬ఆͨ͠ͱ͖ͷ SR ʹΑΔมԽΛਤ 8 ʹࣔ͢ɽද. ͜ΕΑΓɼSumɼLLR (Fixed) ͱൺֱͯ͠ɼఏҊख๏Ͱ. 6ɼਤ 8 ͔Β෼͔ΔΑ͏ʹɼ΄΅͢΂ͯͷۭ࣌ؒղ૾౓ͷ. ͋Δ LLR(GPR) ͷ݁Ռ͕໌Β͔ʹྑ͘ͳ͍ͬͯΔ͜ͱ͕. ૊Ͱɼಛʹ SR ͱ TR ͷͲͪΒ͔ɼ·ͨ͸྆ํ௿͍ͱ͖ʹ. ෼͔Δɽ·ͨɼSR ͱ TR ͷͲͪΒ͔ɼ͋Δ͍͸ͲͪΒ΋. LLR ͷਫ਼౓͕ྑ͘ͳ͍ͬͯΔ͜ͱ͕෼͔Δɽ. ௿͍ͱ͖΋΄΅ LLR (GT) ͱಉ౳ͷਫ਼౓ʹͳ͍ͬͯΔ͜ͱ. ⓒ 2014 Information Processing Society of Japan. 6.

(7) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report ද 7. ͸ SumɼLLR (Fixed)ɼLLR (GPR) ͷதͰ࠷΋ਫ਼౓͕ߴ͍΋ͷΛࣔ͢ɽ 480×360 213×160 128×96 91×68. Fusion SR rule. TR. ςετσʔλͷ͢΂ͯͷۭؒղ૾౓ [ըૉ] ͱ࣌ؒղ૾౓ [fps] ʹ͓͚Δ EER [%]. ଠࣈ. 10. 6. 3.75. 2. 10. 6. 3.75. 2. 10. 6. 3.75. 2. 10. 6. 3.75. 64×48 2. 10. 6. 3.75. 2. Sum. 0.9. 2.5. 8.6. 17.1 1.2 3.4. 9.7. 19.0. 2.8. 6.2. 14.0. 23.6. 5.2. 9.3. 18.9. 29.3. 8.3. 13.8. 24.0. 33.5. LLR (Fixed). 1.0. 2.3. 8.3. 16.9. 1.3. 3.7. 10.0. 19.4. 3.7. 7.7. 15.4. 25.3. 6.7. 11.5. 21.3. 31.8. 9.7. 15.3. 25.7. 35.6. LLR (GPR). 1.0. 2.2. 6.1. 8.3. 1.3. 3.4. 8.7. 12.2 2.3 4.9 12.7 18.0 3.4. 6.8. 15.9 23.2 4.6 10.7 21.5 30.6. LLR (GT). 1.0. 2.2. 5.7. 8.1. 1.1. 3.4. 8.4. 11.8. 6.8. 15.7. (a). (c). 1.9. 4.9. 12.5. 17.3. 3.3. 21.9. 4.4. 10.7. 21.3. 28.7. (b) (a) 30fps. (b) 1fps. (c) 640×480 ըૉ. (d) 53×40 ըૉ. (d). ਤ 7 ۭؒղ૾౓ (ࠨ:480×360 ըૉɼӈ:64×48 ըૉ)ɼ࣌ؒղ૾౓. (্:10fpsɼԼ:2fps) ʹ͓͚Δɼςετσʔλͷ ROC ‫ۂ‬ઢ. ਤ 8. ֶशσʔλʹ͓͚Δ EER ͷɼTR ‫ݻ‬ఆͰͷ SR ʹΑΔมԽ. (্) ͱ SR ‫ݻ‬ఆͰͷ TR ʹΑΔมԽ (Լ)ɽ. ͕෼͔Δɽ ·ͨɼEER ͷ݁ՌΛද 7 ʹࣔ͢ɽද 9 ʹ͸య‫ܕ‬తͳۭ࣌. ͜ͱΛ࣮‫ʹݧ‬ΑΓ֬ೝͨ͠ɽ. ؒղ૾౓Ͱͷ݁ՌΛࣔ͢ɽSR ͱ TR ͷ௿ԼʹΑͬͯ Sum. ࠓ‫ޙ‬ͷ՝୊ͱͯ͠ɼ·ͣɼࠓճ༻͍ͨσʔλ͕ࣨ಺Ͱࡱ. ΍ LLR (Fixed) ͷਫ਼౓͕௿Լ͍ͯͯ͠΋ɼॏΈ෇͚ʹΑͬ. Ө͞Εͨ΋ͷͰ͋Δ͜ͱ͔Βɼ԰֎Λ‫ؚ‬ΊͨΑΓ࣮ࡍతͳ. ͯ LLR (GPR) ͷਫ਼౓͸͋·Γ௿Լ͍ͯ͠ͳ͍͜ͱ͕෼. ‫Ͱڥ؀‬ͷੑೳධՁ͕ඞཁͰ͋Δɽ·ͨɼຊ‫Ͱڀݚ‬͸΄΅ଆ. ͔Δɽ. ໘ํ޲͔ΒࡱӨͨ͠าߦө૾ͷΈΛ༻͍࣮ͯ‫ݧ‬Λߦͬͨ. 5. ͓ΘΓʹ ຊ࿦จͰ͸ɼۭ࣌ؒղ૾౓ʹదԠతͳา༰ɾ‫إ‬ɾ਎௕ʹ. ͕ɼ‫؍‬ଌํ޲ʹΑͬͯา༰ɾ‫إ‬ɾ਎௕ʹର͢ΔॏΈ෇͚͕ มԽ͢Δ͜ͱ΋ߟ͑ΒΕΔͨΊɼ‫؍‬ଌํ޲ͷมԽΛߟྀʹ ೖΕͨॏΈ෇͚ख๏΋ඞཁͰ͋Δɽ. ΑΔϚϧνϞʔμϧ‫ݸ‬ਓೝূͷείΞϨϕϧ౷߹ʹ͍ͭͯ ड़΂ͨɽ࠷ॳʹɼ༷ʑͳۭ࣌ؒղ૾౓ʹ͓͚Δา༰ɾ‫إ‬ɾ. ࢀߟจ‫ݙ‬. ਎௕ͷ֤ϞμϦςΟʹର͢Δେ‫ن‬໛είΞσʔληοτ. [1]. ͷ࡞੒ʹ͍ͭͯઆ໌ͨ͠ɽ࣍ʹɼͦͷσʔληοτΛ༻͍ ͯา༰ɾ‫إ‬ɾ਎௕ͦΕͧΕʹ͍ͭͯ 1 ର 1 ೝূʹΑΔੑೳ. [2]. ධՁΛߦͬͨɽ͞Βʹɼߏஙͨ͠είΞσʔληοτʹ‫ج‬ ͍ͮͯɼۭ࣌ؒղ૾౓ʹదԠతͳείΞ౷߹ํ๏ΛఏҊ͠. [3]. ͨɽ݁Ռͱͯ͠ɼา༰ɼ‫إ‬ɼ਎௕͸ۭ࣌ؒղ૾౓ʹΑͬͯ ड͚ΔӨ‫͕ڹ‬ҟͳΔ͜ͱΛ໌Β͔ʹͨ͠ɽͦͯ͠ɼͦͷ͜. [4]. ͱ͔Βۭ࣌ؒղ૾౓ʹదԠతʹॏΈ෇͚ͷඞཁੑΛ໌Β͔ ʹ͠ɼ࠷దͳॏΈ෇͚Λ͢Δ͜ͱʹΑͬͯਫ਼౓͕޲্͢Δ. ⓒ 2014 Information Processing Society of Japan. [5]. A. K. Jain, P. Flynn, and A. A. Ross, Handbook of Biometrics. Secaucus, NJ, USA: Springer-Verlag New York, Inc., 2007. J. Wambaugh, The Blooding. HarperCollins Publishers, 1989. D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, Handbook of Fingerprint Recognition, 2nd ed. Springer Publishing Company, Incorporated, 2009. D. Zhang, Palmprint Authentication, ser. International Series on Biometrics. Springer Publishing Company, Incorporated, 2004, vol. 3. M. J. Burge and K. W. Bowyer, Handbook of Iris Recog-. 7.

(8) Vol.2014-CVIM-192 No.12 2014/5/15. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. [10]. [11]. (a) 10 fps. (b) 2 fps. [12]. [13]. [14] (c) 480×360 ըૉ. (d) 64×48 ըૉ. ਤ 9 ςετσʔλʹ͓͚Δ EER ͷɼTR ‫ݻ‬ఆͰͷ SR ʹΑΔมԽ. (্) ͱ SR ‫ݻ‬ఆͰͷ TR ʹΑΔมԽ (Լ). [15]. [16]. [17]. (a). (b). [18]. [19]. [20]. ਤ 10. (c) (d) SR (ࠨ: 480×360 ըૉ, ӈ: 64×48 ըૉ) ͱ TR (্: 10. [21]. fps, Լ: 2 fps) ʹ͓͚Δςετσʔλͷ ROC ‫ۂ‬ઢɽ. [6]. [7] [8]. [9]. nition. Springer Publishing Company, Incorporated, 2013. A. K. Jain and S. Z. Li, Handbook of Face Recognition. Secaucus, NJ, USA: Springer-Verlag New York, Inc., 2005. A. Osborn, Questioned Documents, 2nd ed. New York: Boyd Printing Company, 1929. P. S. Teh, A. B. J. Teoh, and S. Yue, “A survey of keystroke dynamics biometrics,” The Scientific World Journal, vol. 2013, no. 408280, pp. 1–24, 2013. A. A. Ross, K. Nandakumar, and A. K. Jain, Handbook. ⓒ 2014 Information Processing Society of Japan. [22]. [23]. of Multibiometrics, ser. Int. Series on Biometrics. Secaucus, NJ, USA: Springer-Verlag New York, Inc., 2006. J. Basak, K. Kate, V. Tyagi, and N. Ratha, “Qplc: A novel multimodal biometric score fusion method,” in IEEE Computer Society and IEEE Biometrics Council Workshop on Biometrics 2010, San Francisco, CA, USA, Jun. 2010, pp. 1–7. J. Fierrez-Aguilar, J. Ortega-Garcia, J. GonzalezRodriguez, and J. Bigun, “Discriminative multimodal biometric authentication based on quality measures,” Pattern Recognition, vol. 38, no. 5, pp. 777–779, May 2005. C. Boehnen, D. Barstow, D. Patlolla, and C. Mann, “A multi-sample standoff multimodal biometric system,” in Biometrics: Theory, Applications and Systems (BTAS), 2012 IEEE Fifth International Conference on, 2012, pp. 127–134. A. Kale, A. Roy-Chowdhury, and R. Chellappa, “Fusion of gait and face for human identification,” in Proc. of the IEEE Int. Conf. on Acoustics, Speech, and Signal Processing 2004 (ICASSP’04), vol. 5, 2004, pp. 901–904. X. Zhou and B. Bhanu, “Feature fusion of side face and gait for video-based human identification,” Pattern Recognition, vol. 41, no. 3, pp. 778–795, 2008. T. Zhang, X. Li, D. Tao, and J. Yang, “Multimodal biometrics using geometry preserving projections,” Pattern Recognition, vol. 41, no. 3, pp. 805–813, 2008. X. Geng, K. Smith-Miles, L. Wang, M. Li, and Q. Wu, “Context-aware fusion: A case study on fusion of gait and face for human identification in video,” Pattern Recogn., vol. 43, no. 10, pp. 3660–3673, Oct. 2010. [Online]. Available: http://dx.doi.org/10.1016/j.patcog.2010.04.012 M. Hofmann, S. M. Schmidt, A. Rajagopalan, and G. Rigoll, “Combined face and gait recognition using alpha matte preprocessing,” in Proc. of the 5th IAPR Int. Conf. on Biometrics, New Delhi, India, Mar. 2012, pp. 1–8. Z.Zhang, “A flexible new technique for camera calibration,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no. 11, pp. 1330–1334, 2000. D. Muramatsu, H. Iwama, Y. Makihara, and Y. Yagi, “Multi-view multi-modal person authentication from a single walking image sequence,” in Biometrics (ICB), 2013 International Conference on, 2013, pp. 1–8. H. Iwama, M. Okumura, Y. Makihara, and Y. Yagi, “The ou-isir gait database comprising the large population dataset and performance evaluation of gait recognition,” IEEE Transactions on Information Forensics and Security, vol. 7, no. 5, pp. 1511–1521, Oct. 2012. J. Han and B. Bhanu, “Individual recognition using gait energy image,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 2, pp. 316– 322, 2006. F. Alonso-Fernandez, J. Fierrez, D. Ramos, and J. Ortega-Garcia, “Dealing with sensor interoperability in multi-biometrics: the upm experience at the biosecure multimodal evaluation 2007,” in Proc. of SPIE 6994, Biometric Technologies for Human Identification IV, Orlando, FL, USA, Mar. 2008. C. K. I. W. Carl Edward Rasmussen, Gaussian Processes for Machine Learning. The MIT Press, 2006.. 8.

(9)