臺灣博碩士論文加值系統

圖形識別已被研究多年，而發展圖形識別的主要目的就是使機器賦有近似人類辨識圖像的能力。然而，由於近幾年的資訊發達，所需處理的文件量亦大為增加，若用人工處理，既費時又費力。因此，有許多學者利用人工智慧的方法來做圖形辨識的應用，以期降低人工處理的成本，其中又以類神經網路被採用的最多。但是對於類神經網路應用於圖形識別的領域中，大多著眼於網路結構或學習方式的改良，並無針對傳統類神經網路應用在圖形辨識的比較與分析。
類神經網路依架構分類，最常見的兩大類分別為前饋式網路及迴遞式網路。本文於這兩大類中，各取三種可用於圖形辨識的類神經網路，分別是倒傳遞網路(BPN)、機率神經網路(PNN)、半徑式函數網路(RBFN)和Hopfield網路、適應共振理論(ART1)網路及雙向聯想記憶(BAM)網路等六種，並以數字為例，將已摻有雜訊或是模糊的數字，用以上六種不同的類神經網路來加以辨識比較，並分析各個網路之雜訊容忍度及其記憶容量。且藉由它們對雜訊的容忍度、記憶容量，以及演算法的不同和各個網路的特性，來加以比較和分析。希望藉由本研究，提供一個與一般論文對類神經網路應用在文字的〝辨識率〞方面，另一種不同的思考方向。

Pattern recognition has been developed for years. The major purpose is to bestow the ability of pattern recognition upon a machine. However, because of the information explosion in recent years, the number of documents that need to be managed has grown rapidly. Because handling documents by hand requires a lot of time and labor, many researchers have applied different artificial intelligence pattern recognition methods to reduce the use of manpower. Among these methods, the neural network is the most widely adopted. However, all of the researches that used the neural network in the field of pattern recognition have involved network structure or learning method improvements. There is no comparison and analysis based on traditional neural network applications to pattern recognition. Neural networks, classified by network structure, could be roughly divided in two categories: feedforward network and recurrent network. This research puts focus on three kinds of neural networks that can be applied to pattern recognition. They are the Backpropagation Network, Probabilistic Neural Network, Radial Basis Function Network, Hopfield Network, Adaptive Resonance Theory Network, and Bidirectional Associative Memory Network. In the digit recognition examples, hazy numerals or cluttered numerals are recognized and compared by these neural networks. The tolerance limit to the clutter and the memory capacity with respect to different neural networks are analyzed. Through the tolerance limit, memory capacity, mathematical calculations, and characteristics of each network, the analysis and the comparison are presented. This study provides another approach for comparison with other researches that focus only on the identification success rate in text recognition.

摘要(中文) Ⅰ
摘要(英文) Ⅱ
致謝 Ⅲ
目錄 Ⅳ
圖目錄 Ⅶ
第一章 緒論 1
1.1 類神經網路的歷史 1
1.2 類神經網路的應用 3
1.3 研究動機 4
1.4 研究目的 4
1.5 論文大綱 5
第二章 類神經網路與文字辨識概述 6
2.1 何謂類神經網路 6
2.2 類神經網路之基本架構 7
2.3 類神經網路之分類 9
2.4 類神經網路的特性 11
2.5 文字辨識概說 12
2.6 相關文獻回顧 13
第三章 前饋式類神經網路數字辨識 21
3.1 倒傳遞類神經網路 21
3.1.1 倒傳遞網路演算法 21
3.1.2 倒傳遞演算法的推導 24
3.1.3 倒傳遞網路的重要參數 28
3.1.4 倒傳遞神經網路的評價 30
3.1.5 模擬結果與討論 31
3.2 機率神經網路 41
3.2.1 機率神經網路演算法 44
3.2.2 機率神經網路的優缺點 45
3.2.3 模擬結果與討論 46
3.3 半徑式函數網路 49
3.3.1 網路架構 50
3.3.2 網路演算法 50
3.3.3 模擬結果與討論 52
第四章 遞迴式類神經網路數字辨識 61
4.1 霍普菲爾網路 61
4.1.1 由能量觀點來看網路的運作過程 61
4.1.2 Hopfield網路演算法之證明 62
4.1.3 Hopfield網路容量問題 65
4.1.4 Hopfield網路之優缺點 65
4.1.5 模擬結果與討論 66
4.2 適應共振理論網路 81
4.2.1 ART1 演算法 82
4.2.2 網路架構 84
4.2.3 ART1的設計原理 85
4.2.4 模擬結果與討論 86
4.3 雙向聯想記憶 92
4.3.1 BAM網路的記憶和回想 92
4.3.2 雙向聯想記憶網路的特性 95
4.3.3 模擬結果與討論 95
第五章 結論與展望 109
參考文獻 111
附件一 115
附件二 116

[1] S. Haykin, NEURAL NETWORKS: A Comprehensive Foundation, Prentice-Hall, Inc.,
1999.
[2] J. A. Anderson, J. W. Silveratein, S. A. Ritz, and R. S. Jones, “Distinctive features, categorical perception, and probability learning: some application of a neural model,” Psychological Review, vol. 84, pp. 413-451, 1977.
[3] T. Kohonen, Self-Organization and Associative memory, Springer-Verlag, Berlin, 1984.
[4] D. Rumelhart, G. Hiton, and J. McClelland, “A General Framework for Parallel Distributed Processing,” Parallel Distributed Processing: Explorations in Microstructure of Cognition, Volume one: Foundations, Bradford Books/MIT Press, Cambridge, MA, pp. 45-76, 1986.
[5] D. Rumelhart and D. Zipser, “Feature Discovery by Competitive Learning,” Parallel Distributed Processing: Explorations in Microstructure of Cognition, Volume one: Foundations., Bradford Books/MIT Press, Cambridge, MA, pp. 151-193, 1986.
[6] G. Hinton and T. Sejnowski, “Learning and Relearning in Boltzmann Machines,” Parallel Distributed Processing: Explorations in Microstructure of Cognition, Volume one: Foundations., Bradford Books/MIT Press, Cambridge, MA, pp. 45-76, 1986.
[7] D. Rumelhart, G. Hinton, and R. Williams, “Learning Internal Representations by Error Propagation,” Parallel Distributed Processing: Explorations in Microstructure of Cognition, Volume one: Foundations., Bradford Books/MIT Press, Cambridge, MA, pp. 318-362, 1986.
[8] B. K. Wong, et al., “A bibliography of neural network business applications research: 1988-September 1994,” Expert Systems, vol. 12, no. 3, pp. 253-262, 1995.
[9] T. Shimada, K. Nishimura, and K. Haruki, “A new self-organizing method and its application to handwritten digit recognition,” Proceedings IEEE International Joint Conference on Neural Networks, vol. 1, pp. 275 —281, 1991.
[10] K. Fukushima, “Neocognitron of a new version: handwritten digit recognition,” Proceedings IEEE International Joint Conference on Neural Networks, vol. 2, pp.1498-1503, 2001.
[11] S. C. Tasty, P. R. Hong, and B. C. Chieu, “Handwritten digits recognition system via OCON neural network by pruning selective update,” Proceedings IEEE International Conference on Pattern Recognition, Conference B: Pattern Recognition Methodology and Systems, vol. 2, pp. 656 —659, 1992.
[12] S. Knerr, L. Personnaz, and G. Dreyfus, “A new approach to the design of neural network classifiers and its application to the automatic recognition of handwritten digits,” Proceedings IEEE International Joint Conference on Neural Networks, vol. 1, pp. 91—96, 1991.
[13] E. Gomez Sanchez, J. A. Gago Gonzalez, Y. A. Dimitriadis, J. M. Cano Izquierdo, and J. Lopez Coronado, “Experimental study of a novel neuro-fuzzy system for on-line handwritten UNIPEN digit recognition,” PATTERN RECOGNITION LETTERS, vol. 19, no. 3-4, pp. 357-364, MAR 1998.
[14] 藍天寶, “Handwritten Characters Recognition Using Two-Class Neural Network and Fuzzy Logic,” M.S. thesis, 中原大學, 1994.
[15] C. A. Perez, G. D. Gonzalez, and C. Salinas, “Genetic selection of non-linear product terms in the inputs to a linear classifier for handwritten digit recognition,” Proceedings IEEE International Conference on Systems, Man, and Cybernetics, vol. 4, pp. 2337—2342, 2001.
[16] 范如逸, “Modified Hopfield Neural Network Based Character Recognition System,” M.S. thesis, 中原大學, 1991.
[17] J. H. Chiang and P. Gader, “A hybrid feature extraction framework for handwritten numeric fields recognition,” Proceedings IEEE International Conference on Pattern Recognition, vol. 3, pp. 436-440, 1996.
[18] 沈緯哲, “CCD機殼序號自動辨識系統,” M.S. thesis, 中央大學, 1999.
[19] 葉怡成, 類神經網路模式應用與實作, 儒林圖書有限公司, 2002.
[20] 王進德, 蕭大全, 類神經網路與模糊控制理論入門, 全華科技圖書股份有限公司, 2001.
[21] H. J. Lee, “Chinese Character Recognition in Taiwan,” 第五屆文字識別與文件分析研討會, 1996.
[22] 戴敏倫, “Neural Network Based Handwritten Alpha-Numeric Recognition System,” M.S. thesis, 中原大學, 1990.
[23] 沈鴻光, “Character Recognition Using Nonlinear Normalization Algorithm and Fuzzy Nearest Prototype Classifier,” M.S. thesis, 中原大學, 1994.
[24] C. Y. Liou and H. C. Yang, “Handprinted Character Recognition Based on Spatial Topology Distance Measurement,” IEEE trans. Pattern Analysis and Machine Intelligence, vol. 16, pp. 394-404, 1994.
[25] C. Y. Liou and H. C. Yang, “Selective Feature-to-Feature Adhesion for Recognition of Cursive Handprinted Characters,” IEEE trans. Pattern Analysis and Machine Intelligence, vol. 21, pp. 184-191, 1999.
[26] 龔純瑩, “The Study of Bidirectional Associative Memory Model and Pattern Recognition,” M.S. thesis, 中原大學, 1991.
[27] W. Deborah, “Selective of Image Primitives for General-Purpose Visual Processing,” Computer Vision, Graphics, and Image Processing, vol. 37, pp. 261-298, 1987.
[28] E. L. Bienenstock, L. N. Cooper, and P. W. Munro, “Theory for the Development of Neuron Selectivity: Orientation Specificity and Binocular Interaction in Visual Cortex,” The Journal of Neuroscience, vol. 2, pp. 32-48, 1982.
[29] 廖建凱, “Pattern Analysis Using Neural Network,” M.S. thesis, 台灣大學, 2001.
[30] C. A. Cruz, Understanding Neural Networks: A Prime, Graeme Publishing Corp., 1988.
[31] J. F. Wang and G. E. Wang, “A New Approach for Recognition of Unconstrained Handwritten Numerals,” proceedings of second National Workshop on Optical Character Recognition, Taipei, Taiwan, Nov. 5, 1992.
[32] D. J. Burr, “Experiments on Neural Net Recognition of Spoken and Written Text,” Processing IEEE Tran. on Acoustics, Speech, and Signal, vol. 36, no. 7, pp. 1162-1168, 1988.
[33] K. Yamada, H. Kami, J. Tsukumo, and T. Temma, “Handwritten Numeral Recognition by Multi-layered Neural Network with Improved Learning Algorithm,” Proceedings, pp. II-259-266, Washington D. C., 1989.
[34] 張森, “A Study on Handwritten Digits Recognition Using Fuzzy r-models and SOM,”
M.S. thesis, 朝陽科技大學, 1999.
[35] 葉怡成, 應用類神經網路, 儒林圖書有限公司, 1999.
[36] J.H. Li, A.N. Michel, and W. Porod, “Analysis and synthesis of a class of neural networks: linear systems operating on a closed hypercube,” Circuits and Systems, IEEE Transactions on , vol. 36, no. 11, pp. 1405 —1422, Nov. 1989.
[37] 羅華強, 類神經網路—MATLAB的應用, 清蔚科技股份有限公司, 2001.
[38] 蘇木春, 張孝德, 機械學習：類神經網路、模糊系統以及基因演算法, 全華科技圖
書股份有限公司, 2000.
[39] B. Kosko, “Bidirectional associative memories,” IEEE Trans. on Systems, Man, and Cybernetics, vol. 18, pp. 49-60, 1988.
[40] Y. F. Wang, J. B. Cruz, and J. H. Mulligan, “Two Coding strategies for bi-directional associative memory,” IEEE Trans. on Neural Networks, vol. 1, no. 1, pp. 81-92, 1990.