IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 MI-Hough Forest () E-mail: ym@vision.cs.chubu.ac.jphf@cs.chubu.ac.jp Abstract Hough Forest Random Forest MI-Hough Forest Multiple Instance Learning Bag Hough Forest Bag Bag Hough Forest 8 7 1 Hough Forest 1 1 [1] SVM[2] Adaboost[3] 08 Leibe Implicit Shape Model(ISM)[4] ISM ISM 09 Gall Random Forest[5] Hough Forest[6] Hough Forest Random Forest 2 Hough Forest 12 Hough Forest [8] [8] ( ) [9] Hough Forest MI-Hough Forest Multiple Instance Learning[7] Bag IS1-09 - 1
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 2 Hough Forest Hough Forest[6] Random Forest( [5]) Random Forest Hough Forest Hough Forest 2.1 Hough Forest 1 2.2 Hough Forest Hough Forest 2 A B Hough Forest 2 Hough Forest 1 Hough Forest 2 Hough Forest MeanShift[10] 3 MI-Hough Forest MI-Hough Forest MI-Hough Forest Multiple Instance Learning[7] Bag 3.1 Bag 3 3 1 I Bag B i = {(I,k, d,o )}(i =1, 2,,I,j = 1, 2,,J) I i Bag j k d o Bag Bag 1 Bag 1 1 Bag 2 IS1-09 - 2
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) (,, ) 3 Bag 3.2 MI-Hough Forest MI-Hough Forest 4 4 MI-Hough Forest Step1 w (d) w (0) =1/N T (t = 1, 2,, T ) 3 T Step2 d h (d) I T S(I,T) τ (1) h (d) T,τ (I) = 0 if S(I,T) < τ (1) 1 otherwise T 5 I T τ Step3 (1) T τ I T 2 (T,τ) U (2) arg min(u ({p i h(i i )=0})+U ({p i h(i i )=1})) T,τ (2) {p i h(i i )=0} {p i h(i i )=1} U 2 A 3 IS1-09 - 3
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 p i Bag (7) p i = 1 p if k =1 (7) J j B i w d+1 (8) p p i w (d+1) p i p if k =1 = otherwise w (d) (8) (9) 5 w (d+1) = w (d+1) w (d+1) i S n j S n k =1 (9) (3) U 1 (A) = A ( c logc (1 c) log(1 c)) (3) c A w (4) c = w (d) i S n j S n k =1 i S n w (d) j S n (4) d (5) U 2 (A) = d A d i:k =1 (d d A ) 2 (5) U 2 Step3 Step4 d Step4 d w p Bag p i p (6) p = 1 1 + exp(1 F(I )) if k = 1 (6) w (d+1) Bag p p Step5Step2 Step4 Step2 Step4 D Step6 L C L D L o 3.3 (1) Dominant Orientation Templates(DOT)[11] DOT DOT 6 0 3 22.5 2 0 3 45 1 0 2 F(I )=2c 1 c p p Bag 4 IS1-09 - 4
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 4. 1 3 (3 ) 5. θ 1 7 6 DOT DOT (10) S(I,T)= P I m I P T m T δ(p I m P T m 0),m=1, 2,,M (10) Pm I P m T I T m DOT AND δ I T AND 1 1 1 3.4 MI-Hough Forest θ 3 y I(y) P (c I(y)) P (c I(y)) θ V θ (y) V θ (y) = P (c I(y)) (11) y I(x) 1. θ 2. 3. θ 4 4.1 ZNCCChamfer MatchingDOT Hough Forest Zero-mean Normalized Cross-Correlation[12] ZNCC Chamfer Matching[13] Chamfer Matching Dominant Orientation Templates[11] DOT 4.2 8 CAD ( 8(a)) ( 8(b)) 3 ( 8 CAD A B C) 0 3 5 IS1-09 - 5
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 9 3 3 9 ZNCCChamfer Matching DOTHough ForestMI-Hough Forest 9(a) B 3pixel 7 3 8 9(b) B 3pixel 6 3 7 4.3 10 cm 3 ( A B C) 1 0 3 3 3 0 0 3pixel 3 11 6 IS1-09 - 6
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 10 ZNCCChamfer MatchingDOTHough Forest MI-Hough Forest 1 B 12 B 13 B 13 B 11 ZNCCChamfer MatchingDOT 3 Hough ForestMI-Hough Forest 3 1 1.1[sec] Hough Forest 1 Hough Forest 1 [1] N. Dalal, and B. Triggs, Histograms of Oriented Gradients for Human Detection, ComputerVisionandPattern Recognition, vol.1, pp.886-893, 05. [2] C. Cortes, and V. Vapnik, Support-vector networks, Machine Learning, pp.273-297,1995. [3] Y. Freund, and R. E. Schapire, Experiments with a new boosting algorithm, InternationalConferenceonMachine Learning, pp.148-156, 1996. [4] B. Leibe, A. Leonardis, and B. Schiele, Robust object detection with interleaved categorization and segmentation, InternationalJournalofComputerVision,vol.77,no.1-3, pp.259-289, 08. [5] L. B. Statistics, and L. Breiman, Random forests, Machine Learning, pp.5-32, 01. [6] J. Gall, and V. Lempitsky, Class-specific hough forests for object detection, Computer Vision and Pattern Recognition, 09. [7] T. G. Dietterich, and R. H. Lathrop, Solving the multipleinstance problem with axis-parallel rectangles, Artificial Intelligence,vol.89, pp.31-71,1997. [8],,,,,, HoughForests, VisionEngineeringWorkshop,12. [9],,,, VisionEngineering Workshop, 12. [10] D. Comaniciu, and P. Meer, Mean Shift Analysis and Applications, in IEEE International Conference on Computer Vision, pp.1197-13, 1999. [11] S. Hinterstoisser, V. Lepetit, S. Ilic, P. Fua, and N. Navab, Dominant orientation templates for real-time detection of texture-less objects,conference Computer Vision and Pattern Recognition, pp.2257-2264,10. [12], http://isl.sist.chukyou.ac.jp/archives/ssii13ts-hashimoto.pdf. [13] H. G.Barrow, J. M.Tenenbaum, R. C.Bolles, and H. C. Wolf, Parametric correspondence and chamfer matching:two new techniques for image matching, Proceedings of the 5th International Joint Conference on Artificial Intelligence, pp.659-663, 1977. 5 MI- 13 7 IS1-09 - 7
IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 12 8 IS1-09 - 8