Lucas-Kanade ヒストグラムマッチングによる対応点を用いない自動色補正 / Auto Color Correction without Correspondence by Lucas-Kanade Histogram Matching slide

Transcript

1. 5511 ' IS1-21: Lucas-Kanade ώετάϥϜϚονϯάʹΑΔ ରԠ఺Λ༻͍ͳ͍ࣗಈ৭ิਖ਼ Auto Color Correction without Correspondence by Lucas-Kanade Histogram Matching দӬ ྗ † Chikara Matsunaga† † גࣜձࣾ๎ӫ ࠤ૔ݚڀ։ൃηϯλʔ † Sakura R&D Center, FOR-A Co., Ltd. E-mail: matsunaga@for-a.co.jp ' ೥ $ ݄  ೔ʢਫʣ
2. ΧϥʔϚονϯά/ΧϥʔΩϟϦϒϨʔγϣϯ ࢀর൘ʢΧϥʔνϟʔτʣΛࡱӨͯ͠ɼΧϝϥؒͷ৭߹ΘͤΛߦ͏1,2,3ɽ 1 দӬ ྗ, ᪅ Ԇ܉, ࿨ా խಙ , ΧϥʔνϟʔτΛ༻͍ͨෳ਺ͷ࠶ࡱϞχλͱΧϝϥͷ࠷ద৭ิਖ਼, ୈ 16 ճը૾ηϯγϯάγϯϙδ΢Ϝ (SSII2010) ߨԋ࿦จू, ԣ඿ (ύγϑΟίԣ඿), 2010 ೥ 6 ݄. 2 দӬ ྗ, 3 ࣍ݩزԿֶม׵ͱزԿֶతϞσϧબ୒ʹΑΔ࠷దΧϥʔϚονϯά/ΧϥʔΩϟϦϒϨʔγϣϯ, ୈ 23 ճը૾ηϯγϯάγϯϙδ΢Ϝ (SSII2017) ߨԋ࿦จू, ԣ඿ (ύγϑΟίԣ඿), 2017 ೥ 6 ݄. 3 দӬ ྗ, ࠷దϨϕϧิਖ਼ͱزԿֶతϞσϧબ୒ʹΑΔߴਫ਼౓৭ิਖ਼ : ը૾ॲཧύΠϓϥΠϯͷߏஙΛ໨ࢦͯ͠, ViEW2017 Ϗδϣϯٕज़ͷ࣮ར༻ϫʔΫγϣοϓߨԋ࿦จू, ԣ඿ (ύγϑΟίԣ඿), 2017 ೥ 12 ݄. দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 2 / 30
3. ҟͳΔࢹ఺͔ΒࡱӨͨ͠ө૾ؒͷ৭߹Θͤ ΧϥʔνϟʔτΛ༻͍ͣʹɼಉҰγʔϯΛҟͳΔࢹ఺͔ΒࡱӨͨ͠ը૾ؒͰΧϥʔϚονϯάΛߦ͏ɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 3 / 30
4. ಛ௃఺Ϛονϯά ORB (Oriented FAST and Rotated BRIEF) 4 ʹΑΔಛ௃఺Ϛονϯάɽ 4 G. Bradski, K. Konolige, V. Rabaud and E. Rublee, ORB: An ecient alternative to SIFT or SURF, 2011 IEEE International Conference on Computer Vision (ICCV 2011), Barcelona, 2011, pp. 25642571. দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 4 / 30
5. ಛ௃఺Ϛονϯάʢଓ͖ʣ (a) (b) RGB ըૉ஋ͷ 3 ࣍ݩϓϩοτྫɽ(a) ը૾தͷ੺࿮ɼ྘࿮಺ͷ RGB ըૉ஋Λ 3 ࣍ݩϓϩοτ͢Δɽ (b) ਖ਼نԽ RGB [0,1] ྖҬʹɼըૉ஋ͱͦΕͧΕͷޡࠩͷପԁମʢ৴པ۠ؒ 95%ʣΛදࣔ͢Δɽ ৭ิਖ਼ͷͨΊʹ͸ɼฏୱͳྖҬʹ͓͚Δ ըૉ஋ Λσʔλͱͯ͠༻͍Δ͜ͱ͕๬·͍͠ɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 5 / 30
6. ݚڀͷ໨త ҟͳΔࢹ఺͔ΒࡱӨͨ͠ը૾ؒͷΧϥʔϚονϯάΛ໨తͱͯ͠ɼըૉ஋ͷ 1 ࣍ݩً౓ɾ 2 ࣍ݩ৭ࠩɼ͋Δ͍͸ 3 ࣍ݩ RGB ώετάϥϜΛϚονϯάͤ͞Δ͜ͱʹΑΓ৭ิਖ਼ͷ ࣗಈԽΛߦ͏ɽ ըૉ஋ώετάϥϜΛը૾ͱݟͳͤ͹ɼը૾ؒͷزԿֶతͳҐஔ߹Θͤʹ༻͍ΒΕΔ Lucas-Kanade ΞϧΰϦζϜΛద༻͢Δ͜ͱ͕Ͱ͖ͯɼ൓෮ຖʹϔοηߦྻΛܭࢉ͠ͳ͍ ٯ݁߹ Lucas-Kanade ΞϧΰϦζϜͷߋ৽Λ 1 ࣍ۙࣅͨۙ͠ࣅٯ݁߹ Lucas-Kanade ΞϧΰϦζϜʹϦʔ୅਺ʹΑΔࢦ਺ࣸ૾Λಋೖ͢Δɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 6 / 30
7. ৭ิਖ਼Ϟσϧ 1 ࣍ݩً౓/2 ࣍ݩ৭ࠩ/3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ʹ͸ɼ ࣍ͷزԿֶม׵ϞσϧΛ༻͍Δɽ 1 ࣍ݩً౓ΞϑΟϯม׵ิਖ਼Ϟσϧ y′ = (1 + p1 )y + p2 . 2 ࣍ݩ৭ࠩΞϑΟϯม׵ิਖ਼Ϟσϧ c′ b = (1 + p1 )cb + p2 cr + p3 , c′ r = p4 cb + (1 + p5 )cr + p6 . 3 ࣍ݩ RGB ΞϑΟϯม׵ิਖ਼Ϟσϧ r′ = (1 + p1 )r + p2 g + p3 b + p4 , g′ = p5 r + (1 + p6 )g + p7 b + p8 , b′ = p9 r + p10 g + (1 + p11 )b + p12 . দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 7 / 30
8. 2 ࣍ݩزԿֶม׵ (a) ฒਐɼ(b) ߶ମɼ(c) ૬ࣅɼ(d) ΞϑΟϯɼ(e) ࣹӨɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 8 / 30
9. 3 ࣍ݩزԿֶม׵ (a) ฒਐɼ(b) ߶ମɼ(c) ૬ࣅɼ(d) ΞϑΟϯɼ(e) ࣹӨɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 9 / 30
10. RGB ً౓৭ࠩม׵ ΨϯϚิਖ਼͞Εͨ RGB ৴߸͔Βً౓৭ࠩ৴߸ Y CB CR ΁ͷม׵͸࣍ͷΑ͏ʹͳΔɽ Y ′ = KR R′ + KG G′ + KB B′, CB = 1 2 B′ − Y ′ 1 − KB , CR = 1 2 R′ − Y ′ 1 − KR . ͜͜ͰɼKR, KG, KB ͸ɼRGB ৭ۭؒʹΑͬͯఆٛ͞Εͨ܎਺Ͱ͋ΓɼKR + KG + KB = 1 Λຬͨ͞ ͳ͚Ε͹ͳΒͳ͍ɽR′, G′, B′ ͸ΨϯϚิਖ਼͞Ε͍ͯΔ͜ͱΛද͢ɽ ITU-R קࠂ BT.709 ن֨5 ͷ৔߹ɼ KB = 0.0722, KR = 0.2126, KG = 1 − KB − KR = 0.7152. 5 Recommendation ITU-R BT.709-6, Parameter values for the HDTV standards for production and international programme exchange (06/2015) দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 10 / 30
11. 1 ࣍ݩً౓ώετάϥϜʢྫʣ 0.0×100 1.0×104 2.0×104 3.0×104 0 10 20 30 40 50 60 Frequency Y দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 11 / 30
12. 2 ࣍ݩ৭ࠩώετάϥϜʢྫʣ -30 -20 -10 0 10 20 30 Cb -30 -20 -10 0 10 20 30 Cr 0.0×100 1.0×104 2.0×104 3.0×104 4.0×104 5.0×104 Frequency দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 12 / 30
13. 1 ࣍ݩ RGB ώετάϥϜʢྫʣ Red Green 0 10 20 30 40 50 60 Blue দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 13 / 30
14. 3 ࣍ݩ RGB ώετάϥϜʢྫʣ ImageJ, An open platform for scientic image analysis, https://imagej.net/Welcome দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 14 / 30
15. ώετάϥϜϚονϯά ೖྗώετάϥϜը૾ hin(x) Λج४ώετάϥϜը૾ href (x) ʹ߹ΘͤΔɽ ͜͜Ͱɼ1 ࣍ݩً౓ώετάϥϜͷ৔߹ɼx = (y, 1)⊤ Ͱ͋Γɼ2 ࣍ݩ৭ࠩώετάϥϜ͸ɼ x = (cb , cr , 1)⊤, 3 ࣍ݩ RGB ώετάϥϜ͸ɼx = (r, g, b, 1)⊤ Ͱ͋ΓɼͦΕͧΕ ըૉ஋ Ͱ͋Δɽ W (x; p) ͸ ώετάϥϜը૾ ͷزԿֶม׵Ͱ͋Γɼp = (p1 , . . . , pn )⊤ ͸ม׵ύϥϝʔλͰ͋Δɽ ྫ͑͹ɼ3 ࣍ݩ RGB ΞϑΟϯม׵ͷ৔߹ɼ W (x; p)=    1 + p1 p2 p3 p4 p5 1 + p6 p7 p8 p9 p10 1 + p11 p12 0 0 0 1       r g b 1    . ͕ͨͬͯ͠ɼج४ώετάϥϜը૾ href ͱج४ώετάϥϜը૾ͷ ըૉ஋ ࠲ඪʹม׵ͨ͠ೖྗ ώετάϥϜը૾ hin ͷؒͷ࣍ͷΑ͏ͳࠩ෼ೋ৐૯࿨ʢSum of Squared Dierence, SSDʣΛ࠷খԽ͢Δɽ ∑ x ( hin(W (x; p))−href (x) )2 . দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 15 / 30
16. Lucas-Kanade ΞϧΰϦζϜ 6,7 Iterate: 1: hin(W (x; p)) Λܭࢉ͢ΔͨΊʹೖྗը૾ hin Λ ύϥϝʔλ W (x; p) Ͱม׵͢Δɽ 2: ޡࠩը૾ href (x) − hin(W (x; p)) Λܭࢉ͢Δɽ 3: ޯ഑ ∇hin Λ W (x; p) Ͱม׵͢Δɽ 4: (x; p) ͰϠίϏߦྻ ∂W ∂p ΛධՁ͢Δɽ 5: ࠷ٸ߱Լը૾ ∇hin ∂W ∂p Λܭࢉ͢Δɽ 6: ࣍ͷϔοηߦྻΛܭࢉ͢Δɽ H = ∑ x ( ∇hin ∂W ∂p )⊤ ( ∇hin ∂W ∂p ) 7: ࣍ͷ࠷ٸ߱Լը૾ͱޡࠩը૾ͷੵ࿨Λܭࢉ͢Δɽ ∑ x ( ∇hin ∂W ∂p )⊤ (href (x) − hin(W (x; p))) 8: ࣍ͷߋ৽ྔ ∆p Λܭࢉ͢Δɽ ∆p = H−1 ∑ x ( ∇hin ∂W ∂p )⊤ (href (x) − hin(W (x; p))) 9: ύϥϝʔλΛ࣍ͷΑ͏ʹߋ৽͢Δɽ p ← p + ∆p Until ∥∆p∥ ≤ εʢඍখ͖͍͠஋ʣ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 16 / 30
17. ٯ݁߹ 7/ۙࣅٯ݁߹ 8,9/ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ΞϧΰϦζϜʢ1ʣ Pre-compute: 3: ج४ը૾ href (x) ͷޯ഑ ∇href ΛධՁ͢Δɽ 4: (x; 0) ͰϠίϏߦྻ ∂W ∂p ΛධՁ͢Δɽ 5: ࠷ٸ߱Լը૾ ∇href ∂W ∂p Λܭࢉ͢Δɽ 6: ࣍ͷϔοηߦྻͷٯߦྻΛܭࢉ͢Δɽ H = ∑ x ( ∇href ∂W ∂p )⊤ ( ∇href ∂W ∂p ) ˌ खॱͷ൪߸͸ɼLucas-Kanade ΞϧΰϦζϜͱରԠ͍ͯ͠Δɽ 6 B. D. Lucas and T. Kanade, An iterative image registration technique with an application to stereo vision, Proceedings of the 7th International Joint Conference on Articial Intelligence - Volume 2 (IJCAI'81), Vancouver, BC, Canada, August 1981, pp. 674679. 7 S. Baker and I. Matthews, Lucas-Kanade 20 years on: A unifying framework, International Journal of Computer Vision, #$-3 (2004), 221255. 8 দӬ ྗ, ը૾͔Βͷഒ཰৭ऩࠩͷࣗಈਪఆิਖ਼, ୈ 20 ճը૾ηϯγϯάγϯϙδ΢Ϝ (SSII2014) ߨԋ࿦จू, ԣ඿ (ύγϑΟίԣ඿), 2014 ೥ 6 ݄. 9 দӬ ྗ, ରԠ఺Λ༻͍ͳ͍ϩʔϦϯάγϟολʔ࿪Έิਖ਼ͱө૾ͷ҆ఆԽ ʙ ฒਐ͔Βճస΁, ୈ 21 ճը૾ηϯγϯάγϯϙδ΢Ϝ (SSII2015) ߨԋ࿦จू, ԣ඿ (ύγϑΟίԣ඿), 2015 ೥ 6 ݄. দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 17 / 30
18. ٯ݁߹ 7/ۙࣅٯ݁߹ 8,9/ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ΞϧΰϦζϜʢ2ʣ Iterate: 1: hin(W (x; p)) Λܭࢉ͢ΔͨΊʹೖྗը૾ hin Λ ύϥϝʔλ W (x; p) Ͱม׵͢Δɽ 2: ޡࠩը૾ hin(W (x; p)) − href (x) Λܭࢉ͢Δɽ 7: ࣍ͷ࠷ٸ߱Լը૾ͱޡࠩը૾ͷੵ࿨Λܭࢉ͢Δɽ ∑ x ( ∇href ∂W ∂p )⊤ (hin(W (x; p)) − href (x)) 8: ࣍ͷߋ৽ྔ ∆p Λܭࢉ͢Δɽ ∆p = H−1 ∑ x ( ∇href ∂W ∂p )⊤ (hin(W (x; p)) − href (x)) 9: ύϥϝʔλΛ࣍ͷΑ͏ʹߋ৽͢Δɽ W (x; p) ← W (x; p) ◦ W (x; ∆p)−1 ʢٯ݁߹ʣ p ← p − ∆p ʢۙࣅٯ݁߹ʣ W (x; p) ← exp G(−∆p)W (x; p) ʢࢦ਺ࣸ૾ٯ݁߹ʣ Until ∥∆p∥ ≤ εʢඍখ͖͍͠஋ʣ ˌ खॱ 9 ͷͦΕͧΕͷ৔߹ͷύϥϝʔλߋ৽ͷࣜͷҧ͍ʹ஫ҙ͢Δɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 18 / 30
19. ࢦ਺ࣸ૾ ແݶখม׵ͷϦʔ୅਺͸ม׵܈ Mp ͷ p ʹ͓͚Δ઀ۭؒ Tp(Mp) ͱݟͳͤΔɽϦʔ୅਺ʹ͓͍ͯܭࢉͨ͠૿෼ ∆p Λࢦ਺ؔ਺ exp G(∆p) ʹΑͬͯɼྡ઀͢Δม׵܈ Mp ͷ఺ʹࣹӨ͢Δ10ɽ͜͜Ͱɼexp G(∆p) ͸Ϧʔ୅਺ͷࢦ਺ࣸ૾Ͱ͋Γɼྫ͑͹ɼ3 ࣍ ݩ RGB ΞϑΟϯม׵ͷ৔߹ʹ͸ɼ࣍ͷΑ͏ʹͳΔɽ exp G(∆p) = exp ( 12 ∑ n=1 ∆pnGn ) = I4×4 + 12 ∑ n=1 ∆pnGn + 1 2 ( 12 ∑ n=1 ∆pnGn )2 + · · · I4×4 ͸ 4 ࣍୯ҐߦྻͰ͋ΔɽGn ͸ແݶখม׵ͷੜ੒ࢠͰ͋Δɽ 10 K. Kanatani, Lie algebra method for pose optimization computation, In: O. Sergiyenko, W. FloresFuentes, P. Mercorelli (eds.), Machine Vision and Navigation, Springer, 2019, to appear. দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 19 / 30
20. Lucas-Kanade ώετάϥϜϚονϯάॲཧ Lucas-Kanade ώετάϥϜϚονϯάॲཧશମϒϩοΫਤɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 20 / 30
21. Reinhard άϩʔόϧ TM 11 0.0 0.2 0.4 0.6 0.8 1.0 0 1 2 3 4 5 Lwhite = 0.5 Lwhite = 1.0 Lwhite = 1.5 Lwhite = 3.0 Lwhite = inf World luminance (L) Ld Ld (x, y) = L(x, y) ( 1 + L(x, y)/L2 white ) 1 + L(x, y) . 11 E. Reinhard, G. Ward, S. Pattanaik, and P. Debevec, High Dynamic Range Imaging: Acquisition, Display, and Image-based Lighting, Amsterdam, Elsevier/Morgan Kaufmann, 2005. দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 21 / 30
22. ΧϥʔϚονϯάධՁํ๏ ධՁ༻ը૾ͱͯ͠ɼ ಉҰը૾12 ΛࣹӨม׵͢Δ͜ͱʹΑΓੜ੒ٖͨ͠ࣅଟࢹ఺ը૾ ࿈ଓ͢Δಈը૾ྻ͔Βͷ 2 ը૾13 εςϨΦը૾12 Λ༻͍ͯɼͦΕͧΕɼ3 ࣍ݩ RGB ৭ۭؒʹ͓͚Δ૬ࣅม׵ʹΑΓ৭ม׵ͨ͠ը૾ͷج४ը૾ʹର͢Δ ৭ิਖ਼ύϥϝʔλΛਪఆͯ͠ɼ৭ิਖ਼͢Δɽͦͯ͠ɼਅͷը૾ͱ৭ิਖ਼ը૾ͷؒͷ৭ࠩ ∆E∗ ab ʹΑΓ ධՁ͢Δɽ৭ࠩࣜʹ͸ CIEDE2000 (∆E∗ 00) 14 Λ༻͍ͨɽ ֤ධՁ༻ը૾ 10 ૊Λɼ૬ࣅม׵ύϥϝʔλΛม͑ͯɼ1 ૊ʹ͖ͭ 100 ճߦͬͨɽٖࣅଟࢹ఺ը૾ͷ৔߹ ͸ɼਖ਼نཚ਺ʹΑΓࣹӨม׵΋ม͍͑ͯΔɽ 12 Middlebury Stereo Datasets, http://vision.middlebury.edu/stereo/data/ 13 Video Quality Experts Group (VQEG), https://www.its.bldrdoc.gov/vqeg/downloads.aspx 14 ೔ຊ޻ۀن֨ JIS Z8730:2009 ৭ͷදࣔํ๏  ෺ମ৭ͷ৭ࠩ Colour specication - Colour dierences of object colours দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 22 / 30
23. ΧϥʔϚονϯάධՁ݁Ռʢ1ʣ ٖࣅଟࢹ఺ը૾ͷ৔߹ɽݪը૾ͱ৭ิਖ਼ը૾ͷؒͷ৭ࠩ ∆E∗ ab ͷฏۉɼฏۉऩଋճ਺ ʢً౓৭ࠩ৭ۭؒͷ৔߹͸ 2 ࣍ݩ৭ࠩώετάϥϜͷ݁Ռʣ ɽׅހ಺͸ඪ४ภࠩͰ͋Δɽ ৭ۭؒ ํ๏ ฏۉ৭ࠩ ฏۉऩଋճ਺ ً౓৭ࠩ Lucas-Kanade ώετάϥϜϚονϯά 2.152 (±2.528) 5.03 (±1.25) ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 2.255 (±2.510) 4.73 (±1.10) ۙࣅٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 2.313 (±3.304) 4.81 (±1.05) ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 2.114 (±1.357) 5.85 (±1.49) RGB Lucas-Kanade ώετάϥϜϚονϯά 0.474 (±0.330) 6.70 (±1.93) ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 0.473 (±0.324) 6.21 (±1.71) ۙࣅٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 0.473 (±0.324) 6.52 (±1.88) ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 0.473 (±0.324) 7.41 (±2.23) RGB ฏۉඪ४ภࠩϚονϯά 1.187 (±0.467)  দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 23 / 30
24. ΧϥʔϚονϯάධՁ݁Ռʢ2ʣ ಈը૾ྻ͔Βͷ 2 ը૾ͷ৔߹ɽݪը૾ͱ৭ิਖ਼ը૾ͷؒͷ৭ࠩ ∆E∗ ab ͷฏۉɼฏۉऩଋճ ਺ʢً౓৭ࠩ৭ۭؒͷ৔߹͸ 2 ࣍ݩ৭ࠩώετάϥϜͷ݁Ռʣ ɽׅހ಺͸ඪ४ภࠩͰ͋Δɽ ৭ۭؒ ํ๏ ฏۉ৭ࠩ ฏۉऩଋճ਺ ً౓৭ࠩ Lucas-Kanade ώετάϥϜϚονϯά 1.645 (±0.899) 4.70 (±0.84) ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 1.629 (±0.891) 4.38 (±0.80) ۙࣅٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 1.629 (±0.891) 4.51 (±0.87) ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 1.629 (±0.891) 5.48 (±1.34) RGB Lucas-Kanade ώετάϥϜϚονϯά 0.697 (±0.306) 6.64 (±1.36) ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 0.703 (±0.300) 6.51 (±1.17) ۙࣅٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 0.703 (±0.300) 6.68 (±1.18) ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 0.703 (±0.300) 7.57 (±1.75) RGB ฏۉඪ४ภࠩϚονϯά 1.310 (±0.583)  দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 24 / 30
25. ΧϥʔϚονϯάධՁ݁Ռʢ3ʣ εςϨΦը૾ͷ৔߹ɽݪը૾ͱ৭ิਖ਼ը૾ͷؒͷ৭ࠩ ∆E∗ ab ͷฏۉɼฏۉऩଋճ਺ ʢً౓৭ࠩ৭ۭؒͷ৔߹͸ 2 ࣍ݩ৭ࠩώετάϥϜͷ݁Ռʣ ɽׅހ಺͸ඪ४ภࠩͰ͋Δɽ ৭ۭؒ ํ๏ ฏۉ৭ࠩ ฏۉऩଋճ਺ ً౓৭ࠩ Lucas-Kanade ώετάϥϜϚονϯά 2.693 (±1.427) 5.41 (±1.46) ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 2.790 (±2.559) 5.25 (±1.12) ۙࣅٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 2.710 (±1.418) 5.37 (±1.13) ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 2.705 (±1.419) 6.55 (±2.60) RGB Lucas-Kanade ώετάϥϜϚονϯά 1.122 (±0.531) 9.24 (±3.50) ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 1.172 (±0.588) 8.94 (±2.48) ۙࣅٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 1.172 (±0.588) 9.17 (±2.50) ࢦ਺ࣸ૾ٯ݁߹ Lucas-Kanade ώετάϥϜϚονϯά 1.172 (±0.588) 9.60 (±2.84) RGB ฏۉඪ४ภࠩϚονϯά 1.754 (±0.652)  দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 25 / 30
26. ৭ͷڐ༰ࠩͷࣄྫ ৭ࠩͷେ͖͞ ໊ শ ఠ ཁ ∼ 0.2 ଌ৭ෆೳྖҬ 0.3 ࣝผ৭ࠩ ಉҰ෺ମͷଌ৭࠶ݱਫ਼౓ 0.6 1 ڃʢݫ֨৭ࠩʣ ֤छͷޡࠩཁҼΛߟ͑ͨ৔߹ͷ࣮༻తͳڐ༰ࠩͷݶք 1.2 2 ڃʢ࣮༻৭ࠩ aʣ ฒ΂ͯ൑ఆͨ͠৔߹ʹɼ΄ͱΜͲͷਓ͕༰қʹ৭ࠩΛೝΊΔ͜ͱ͕Ͱ͖Δɽ Ϛϯηϧ AAA ڃɼ๷Ӵி OD ৭ 2.5 3 ڃʢ࣮༻৭ࠩ bʣ ཭ؒͯ͠൑ఆͨ͠৔߹ɼ΄΅ಉҰͱೝΊΔ͜ͱ͕Ͱ͖Δɽ Ϛϯηϧ AA ڃɼJIS ඪ४৭ථ 5.0 4 ڃ ܦ࣌ൺֱͨ͠৔߹ʹɼ΄΅ಉҰͱೝΊΔ͜ͱ͕Ͱ͖Δɽ Ϛϯηϧ A ڃ 10.0 5 ڃ ϚʔΩϯάϖϯʢJIS S6037-1992ʣ 20.0 6 ڃ ৭໊Ϩϕϧͷ৭ͷ؅ཧ ࣗಈं෦඼ͷృບ௨ଇʢJIS D0202-1988ʣ ౿੾ॾࢪઃͷ৭࠼ʢJIS E3701-1984ʣ ελϯϓ୆ʢJIS S6016-1991ʣ Ԗචɼ৭Ԗචٴͼγϟʔϓϖϯγϧʹ༻͍Δ͠ΜʢJIS S6005-1992ʣ ҆શ৭࠼࢖༻௨ଇʢJIS Z9101-1986ʣ ഑؅ܥͷࣝผදࣔʢJIS Z9102-1987ʣ ߤۭӉ஦-഑؅-ࣝผʢJIS W0601-1990ʣ ˌ ೔ຊి৭޻ۀגࣜձࣾϗʔϜϖʔδΑΓൈਮ. দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 26 / 30
27. ΧϥʔϚονϯάը૾ྫʢ1ʣ -100 0 100 200 300 -100 0 100 200 300 -100 0 100 200 300 B Reference Input R G B -100 0 100 200 300 -100 0 100 200 300 -100 0 100 200 300 B Reference Correct R G B (a) (b) (c) (d) 10-4 10-3 10-2 10-1 1 3 5 7 9 Iteration Residual FA IC Approx.IC expM Red Conv RGB LK Green Conv RGB LK 110 120 130 140 150 160 Blue Pixel val Conv RGB LK (e) (f) (g) (h) ٖࣅଟࢹ఺ը૾ͷ৔߹ɽ(a) ج४ը૾ͱͦͷࣹӨม׵ʹΑΔೖྗݪը૾ (b)ɼ(c) ج४ը૾ʢReferenceʣͱೖྗݪը૾ͷ৭ม׵ը૾ ʢInputʣͷ RGB ըૉ஋ͷ 3 ࣍ݩϓϩοτɼ(d) ج४ը૾ͱ 3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ը૾ʢCorrectʣͷ RGB ըૉ஋ͷ 3 ࣍ݩϓϩοτɼ(e) ೖྗݪը૾ͷ৭ม׵ը૾ɼ(f) 3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ը૾ɼ(g) 3 ࣍ ݩ RGB ώετάϥϜϚονϯάʹ͓͚Δ൓෮ճ਺ʹର͢Δ࢒ࠩάϥϑɼ(h) ೖྗݪը૾ͱ৭ม׵/৭ิਖ਼ը૾ͷؒͷࠩ෼Φϑηοτը ૾ͷ RGB ຖͷώετάϥϜɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 27 / 30
28. ΧϥʔϚονϯάը૾ྫʢ2ʣ -100 0 100 200 300 -100 0 100 200 300 -100 0 100 200 300 B Reference Input R G B -100 0 100 200 300 -100 0 100 200 300 -100 0 100 200 300 B Reference Correct R G B (a) (b) (c) (d) 10-4 10-3 10-2 10-1 1 3 5 7 9 Iteration Residual FA IC Approx.IC expM Red Conv RGB LK Green Conv RGB LK 110 120 130 140 150 160 Blue Pixel val Conv RGB LK (e) (f) (g) (h) ಈը૾ྻ͔Βͷ 2 ը૾ͷ৔߹ɽ(a) ج४ը૾ͱೖྗݪը૾ (b)ɼ(c) ج४ը૾ʢReferenceʣͱೖྗݪը૾ͷ৭ม׵ը૾ʢInputʣͷ RGB ըૉ஋ͷ 3 ࣍ݩϓϩοτɼ(d) ج४ը૾ͱ 3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ը૾ʢCorrectʣͷ RGB ըૉ஋ ͷ 3 ࣍ݩϓϩοτɼ(e) ೖྗݪը૾ͷ৭ม׵ը૾ɼ(f) 3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ը૾ɼ(g) 3 ࣍ݩ RGB ώε τάϥϜϚονϯάʹ͓͚Δ൓෮ճ਺ʹର͢Δ࢒ࠩάϥϑɼ(h) ೖྗݪը૾ͱ৭ม׵/৭ิਖ਼ը૾ͷؒͷࠩ෼Φϑηοτը૾ͷ RGB ຖ ͷώετάϥϜɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 28 / 30
29. ΧϥʔϚονϯάը૾ྫʢ3ʣ -100 0 100 200 300 -100 0 100 200 300 -100 0 100 200 300 B Reference Input R G B -100 0 100 200 300 -100 0 100 200 300 -100 0 100 200 300 B Reference Correct R G B (a) (b) (c) (d) 10-4 10-3 10-2 10-1 1 4 7 10 13 Iteration Residual FA IC Approx.IC expM Red Conv RGB LK Green Conv RGB LK 110 120 130 140 150 160 Blue Pixel val Conv RGB LK (e) (f) (g) (h) εςϨΦը૾ͷ৔߹ɽ(a) ج४ը૾ͱೖྗݪը૾ (b)ɼ(c) ج४ը૾ʢReferenceʣͱೖྗݪը૾ͷ৭ม׵ը૾ʢInputʣͷ RGB ըૉ஋ͷ 3 ࣍ݩϓϩοτɼ(d) ج४ը૾ͱ 3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ը૾ʢCorrectʣͷ RGB ըૉ஋ͷ 3 ࣍ݩϓϩο τɼ(e) ೖྗݪը૾ͷ৭ม׵ը૾ɼ(f) 3 ࣍ݩ RGB ώετάϥϜϚονϯάʹΑΔ৭ิਖ਼ը૾ɼ(g) 3 ࣍ݩ RGB ώετάϥϜϚονϯ άʹ͓͚Δ൓෮ճ਺ʹର͢Δ࢒ࠩάϥϑɼ(h) ೖྗݪը૾ͱ৭ม׵/৭ิਖ਼ը૾ͷؒͷࠩ෼Φϑηοτը૾ͷ RGB ຖͷώετάϥϜɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 29 / 30
30. ·ͱΊ ҟͳΔࢹ఺͔ΒࡱӨͨ͠ը૾ؒͷΧϥʔϚονϯάΛ໨తͱͯ͠ɼըૉ஋ͷ 1 ࣍ݩً౓ɾ 2 ࣍ݩ৭ࠩɼ͋Δ͍͸ 3 ࣍ݩ RGB ώετάϥϜΛϚονϯάͤ͞Δ͜ͱʹΑΓ৭ิਖ਼ͷ ࣗಈԽΛߦͬͨɽ ըૉ஋ώετάϥϜΛը૾ͱݟͳͤ͹ɼը૾ؒͷزԿֶతͳҐஔ߹Θͤʹ༻͍ΒΕΔ Lucas-Kanade ΞϧΰϦζϜΛద༻͢Δ͜ͱ͕Ͱ͖ͯɼ൓෮ຖʹϔοηߦྻΛܭࢉ͠ͳ͍ ٯ݁߹ Lucas-Kanade ΞϧΰϦζϜͷߋ৽Λ 1 ࣍ۙࣅͨۙ͠ࣅٯ݁߹ Lucas-Kanade ΞϧΰϦζϜʹϦʔ୅਺ʹΑΔࢦ਺ࣸ૾Λಋೖͨ͠ɽ ධՁ༻ը૾ͱͯ͠ɼ ಉҰը૾ΛࣹӨม׵͢Δ͜ͱʹΑΓੜ੒ٖͨ͠ࣅଟࢹ఺ը૾ ࿈ଓ͢Δಈը૾ྻ͔Βͷ 2 ը૾ εςϨΦը૾ Λ༻͍ͯɼͦΕͧΕɼ3 ࣍ݩ RGB ৭ۭؒʹ͓͚Δ૬ࣅม׵ʹΑΓ৭ม׵ͨ͠ը૾ͷج४ը૾ ʹର͢Δ৭ิਖ਼ύϥϝʔλΛਪఆͯ͠ɼ৭ิਖ਼ͨ͠ɽਅͷը૾ͱ৭ิਖ਼ը૾ͷؒͷ৭ࠩ ∆E∗ ab ʹΑΓධՁͨ͠ɽ ͢΂ͯͷॲཧ͸ɼώετάϥϜಛ௃ۭؒʹ͓͚ΔϚονϯάͰ͋Γɼը૾ؒͷزԿֶతͳ Ґஔ߹ΘͤΛߦΘͳ͍͚ͩͰͳ͘ɼԿΒͷಛ௃఺ͷநग़΍ରԠ෇͚΋ඞཁͱ͠ͳ͍ɽ দӬ ྗ (๎ӫ) Lucas-Kanade ώετάϥϜϚονϯά 2019 ೥ 6 ݄ 12 ೔ʢਫʣ 30 / 30