國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
第七章 結論與後續研究改進方向
本研究的目標是改善區域二元化圖型的抗噪性與描述性，期望應用於物件分類或 辨識時能有較佳的效能與準確性。我們將新提出的描述子並命名為延展式區域三元化 圖型。在定義 ELTP 時，對於轉換為三元編碼後導致樣式種類大量增的問題，我們利 用了 Spectral clustering 有效的解決了這個問題，並且成功的提升了 ELTP 的抗噪能力。
對於 LBP 中 Uniform pattern 的存在，我們也在 ELTP 中找出三元編碼的 Uniform pattern，並對三元編碼的 Uniform pattern 數量過多的問題，有效的使用了 Spectral clustering 來解決。在第五章的實驗中，我們驗證了 ELTP 的抗噪能力，而在第六章的 圖型識別應用中，不論是材質分析或是人臉辨識，ELTP 都出色的結果。
對於 UELTP1 的 Uniform pattern 來說，其 pattern 數只佔了 2.5%，但是卻可在圖 像中佔到 20~40%。對於這種特性若使用分群將其聚合，似乎有些可惜，因此未來在 對於三元編碼的 Uniform pattern 的定義與降維，可對 UELTP1 做更進一步的研究，看 其能否更進一步的增強抗噪力與描述力。
此外，本研究所提出的 ELTP 乃基於原始 LBP 的定義，但是 LBP 本身也有許多的 變形，如 Rotation-invariant LBP[8]、XY-LBP[19]、volume LBP (VLBP)[20]、CS-LBP[21]
等，如何將本論文所提出三元編碼的概念套用至上述的描述子中，將是值得後續研究 探討的問題。
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
參考文獻
[1] P. Viola and M. Jones, “Robust Real-Time Object Detection”. Proc. ICCV Second Int'l Workshop Statistical and Computational Theories of Vision Modelling, Learning, Computing, and Sampling, July 2001.
[2] N. Dalal and B. Triggs, “Histograms of Oriented Gradients for Human Detection”, Proc.
IEEE Conf. Computer Vision and Pattern Recognition (CVPR '05), vol. 1, pp. 886-893, 2005.
[3] R.Plamondon and S.N. Srihari, “On-Line and Off-Line Handwriting Recognition: A Comprehensive Survey”. IEEE Trans. Pattern Analysis and Machine Intelligence, vol.
22, no. 1, pp. 63-84, Jan. 2000 [4] Google goggles
www.google.com/mobile/goggles [5] VOC 2009 Challenge Results:
http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2009/results/index.html
[6] D. Lowe, “Distinctive Image Features from Scale-Invariant Keypoints,” Int'l J.
Computer Vision, vol. 2, no. 60, pp. 91-110, 2004.
[7] B. S. Manjunath, J.-R. Ohm, V. V. Vasudevan, and A. Yamada ,“Color and texture descriptors,” IEEE Trans. Circuit Syst. Video Technol., vol. 11, pp. 703–715, June 2001
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
[8] T. Ojala, M. Pietikainen, and T. Maenpaa, “Multiresolution Gray-Scale and Rotation Invariant Texture Classification with Local Binary Patterns,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 24, no. 7, pp. 971-987, July 2002.
[9] Pontil and A. Verri, “Support Vector Machines for 3D Object Recognition”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 20(6), pp. 637-646, 1998.
[10] A. Gionis, P. Indyk, and R. Motwani, “Similarity Search in High Dimensions via Hashing,” Proc. Very Large Data Base Conf. (VLDB '99), pp. 518–529, Sept. 1999.
[11] T. Maenpaa, and M. Pictikainen, “Multi-scale binary patterns for texture analysis,”
Springer Berlin / Heidelberg, 2003.
[12]C. He, T. Ahonen and M. Pietikäinen, “A Bayesian Local Binary Pattern texture descriptor”,Proc. Int’ l Conf. on Pattern Recognition, 2008.
[13] X. Tan and B. Triggs. “Enhanced local texture feature sets for face recognition under difficult lighting conditions”. In Analysis and Modeling of Faces and Gestures, volume 4778 of LNCS, pages 168–182. Springer, 2007
[14] Matthias Hein and Ulrike von Luxburg ,“Short Introduction to Spectral Clustering”, MLSS 2007
[15] Ng, A., Jordan, M., and Weiss, Y. (2002). On spectral clustering: analysis and an algorithm. In T. Dietterich,S. Becker, and Z. Ghahramani (Eds.), Advances in Neural Information Processing Systems 14 (pp. 849 –856). MIT Press.
[16] Brodatz database
http://www.ux.uis.no/~tranden/
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
[17] T. Ahonen, A. Hadid, and M. Pietikainen, “Face Description with Local Binary Patterns:
Application to Face Recognition,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 28, no. 12, pp.2037-2041, Dec. 2006.
[18] The Yale Face Database B
http://cvc.yale.edu/projects/yalefacesB/yalefacesB.html
[19] G. Zhao and M. Pietik¨ainen. Dynamic texture recognition using local binary patterns with an application to facial expressions. IEEE Trans. PAMI, 29(6):915–928, 2007.
[20] G.Zhao and M. Pietikäinen, “Dynamic Texture Recognition Using Volume Local Binary Patterns”, Proc. ECCV 2006 Workshop on Dynamical Vision, Graz, Austria, 2006, accepted.
[21] M. Heikkil¨a, M. Pietik¨ainen, and C. Schmid, “Description of interest regions with center-symmetric local binary patterns”,In Computer Vision, Graphics and Image Processing, 5th Indian Conference, pages 58–69, 2006.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
附錄 A 材質分類實驗結果(二)數據
最小閾值：2 雜訊強度： SNR=33
樣本大小為 64×64 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 98.71 98.62 98.11 98.8
Noisy 80.57 77.2 79.37 78.73
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 98.8 99.49 99.08 99.36
Noisy 85.27 87.29 86.97 85.27
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.17 99.49 99.13 99.59
Noisy 84.21 83.75 82 84.49
樣本大小為 96×96 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.54 99.28 99.13 99.59
Noisy 82 81.43 81.02 83.18
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.54 99.69 99.54 99.69
Noisy 88.84 89.15 90.07 89.56
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.53 99.85 99.59 99.74
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
樣本大小為 128×128 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.31 99.48 99.37 99.77
Noisy 80.57 82.36 81.21 87.01
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.48 99.83 99.54 99.66
Noisy 90 91.03 90.92 89.77
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.42 99.89 99.43 99.89
Noisy 84.65 88.39 87.7 85.63
雜訊強度：SNR=25
樣本大小為 64×64 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 98.71 98.62 98.11 98.8
Noisy 69.66 66.71 68.05 68.05
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 98.8 99.49 99.08 99.36
Noisy 71.18 77.12 76.52 72.84
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.17 99.49 99.13 99.59
Noisy 75.32 74.13 67.91 78.91
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
樣本大小為 96×96 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.54 99.28 99.13 99.59
Noisy 75.51 73.15 71.97 75.51
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.54 99.69 99.54 99.69
Noisy 77.21 85.24 83.23 81.33
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.54 99.85 99.59 99.74
Noisy 78.18 85.19 80.09 82.97
樣本大小為 128×128 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.31 99.48 99.37 99.77
Noisy 75.69 75.34 73.44 78.74
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.48 99.83 99.54 99.66
Noisy 78.39 85.69 88.16 82.01
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.42 99.89 99.43 99.89
Noisy 74.77 85.06 80.63 83.45
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
雜訊強度：SNR=20
樣本大小為 64×64 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 98.71 98.62 98.11 98.8
Noisy 57.87 56.22 53.41 48.99
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 98.8 99.49 99.08 99.36
Noisy 53.41 66.25 61.6 52.95
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.17 99.49 99.13 99.59
Noisy 59.43 57.69 52.3 62.11
樣本大小為 96×96 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.54 99.28 99.13 99.59
Noisy 67.13 64.09 62.14 65.23
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.54 99.69 99.54 99.69
Noisy 60.24 75.46 72.84 63.43
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.53 99.85 99.59 99.74
Noisy 68.52 75.67 66.98 72.53
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
樣本大小為 128×128 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.31 99.48 99.37 99.77
Noisy 70.51 67.76 66.83 66.95
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.48 99.83 99.54 99.66
Noisy 62.47 79.14 76.32 66.72
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.42 99.89 99.43 99.89
Noisy 68.04 77.24 69.83 74.38
最小閾值：5
雜訊強度： SNR=33
樣本大小為 64×64 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 98.66 98.16 97.65 99.08
Noisy 74.17 69.1 74.22 68.64
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 98.71 99.59 99.13 99.49
Noisy 74.72 80.43 79.42 79.6
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.94 99.45 99.03 99.36
Noisy 78.13 77.16 77.67 79.47
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
樣本大小為 96×96 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.58 99.28 99.17 99.64
Noisy 76.33 71.81 77.57 73.71
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.54 99.74 99.64 99.69
Noisy 79.78 82.56 81.64 80.76
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.54 99.79 99.64 99.69
Noisy 81.33 81.12 81.74 81.12
樣本大小為 128×128 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.31 99.42 99.31 99.77
Noisy 79.2 70.11 75.59 76.55
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.31 99.89 99.48 99.6
Noisy 80.63 85.06 82.36 83.22
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.37 99.83 99.48 99.77
Noisy 80.63 81.03 82.82 82.36
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
雜訊強度：SNR=25
樣本大小為 64×64 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.67 98.16 97.65 99.08
Noisy 65.79 62.43 62.43 59.02
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 98.71 99.59 99.13 99.49
Noisy 62.34 69.48 69.8 67.86
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 98.94 99.45 99.03 99.36
Noisy 67.77 67.73 66.71 72.15
樣本大小為 96×96 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.59 99.28 99.18 99.64
Noisy 72.53 68.78 68.05 67.49
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.54 99.74 99.64 99.69
Noisy 68.11 79.37 75.51 73.41
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.54 99.79 99.64 99.69
Noisy 74.13 76.03 74.95 76.75
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
樣本大小為 128×128 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.31 99.43 99.31 99.77
Noisy 75.28 69.14 71.09 70.74
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.31 99.89 99.48 99.6
Noisy 69.66 79.77 77.76 74.89
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.37 99.83 99.48 99.77
Noisy 76.09 75.63 78.16 79.2
雜訊強度：SNR=20
樣本大小為 64×64 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 98.66 98.16 97.65 99.08
Noisy 54.6 54.88 53.22 44.94
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 98.71 99.59 99.13 99.49
Noisy 50.46 60.08 59.71 53.91
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 98.94 99.45 99.03 99.36
Noisy 56.49 56.72 51.8 58.7
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
樣本大小為 96×96 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.59 99.28 97.18 99.64
Noisy 61.99 63.89 60.55 58.74
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.54 99.74 99.64 99.69
Noisy 56.68 68.06 59.55 62.6
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.5 99.79 99.64 99.69
Noisy 66.35 70.68 66.2 69.55
樣本大小為 128×128 的實驗結果(準確率%)
Gaussian64 Gaussian128 Hamming64 Hamming128
Original 99.31 99.43 99.31 99.77
Noisy 69.48 66.55 63.39 65.23
UELTP3-G58 UELTP3-G128 UELTP3-H58 UELTP3-H128
Original 99.31 99.89 99.48 99.6
Noisy 60.23 74.66 72.01 66.26
UELTP4-G58 UELTP4-G128 UELTP4-H58 UELTP4-H128
Original 99.36 99.83 99.48 99.77
Noisy 67.35 72.3 68.85 74.94