隱藏式馬可夫模型是一組統計模型用來描述統計學上具有狀態轉移的隨機程序訊號，這種描述能力已被廣泛的應用在語音辨識上。由於語音辨識常會發生時間校準的問題，這種模型可運用狀態轉移的觀念解決這種問題。在隱藏式馬可夫模型訓練的階段時，利用Viterbi搜尋演算法預測出一組最適合的狀態序列來描述訓練語料。觀察樣本特徵將根據所對應的最佳序列來分配至各個狀態中並訓練出模型所需的各項參數。因此如何確保觀察樣本特徵被正確的分配到其所對應的狀態實為影響模型訓練正確性的重要因素。本論文從假說檢定為出發點推導出最小化驗證錯誤與最大化信賴量測之關係以獲得新的輔助函式。此輔助函式在Viterbi搜尋過程中不斷解碼出最佳狀態序列並作為分配訓練狀態模型所需觀察樣本的依據。不僅如此，在輔助函式裡我們也導入一個鑑別性轉換矩陣，在最大化輔助函式同時將轉換矩陣求出。我們證明在大資料量與描述各狀態的共變異矩陣相等的假設下，此轉移矩陣會等同於實現傳統線性鑑別式分析。此轉換矩陣可以用來擷取出具鑑別性且低維度的特徵資料，使得資料在每一個訓練回合可以更正確的分配到其對應的狀態。我們將這樣的成果運用在可以描述二維空間人臉影像資料之嵌入式隱藏式馬可夫模型訓練上，以克服因人臉角度不同造成人臉影像在平面空間上垂直水平歪斜問題。用訓練出的鑑別性隱藏式馬可夫狀態模型有效進行空間上的校準。本論文採用ORL，FERET 及UMIST三套人臉資料庫來評估本演算法的辨識效能並與傳統嵌入式隱藏式馬可夫模型來做比較，此外我們也評估鑑別式轉換矩陣所擷取資料維度對人臉辨識率的影響。實驗結果可以發現當適合的資料維度被決定，我們所擷取出來的具鑑別性資料搭配上修改後的Viterbi搜尋演算法可以有較高的人臉辨識率及較快的辨識速度。

　　Hidden Markov Model (HMM) is feasible to describe stochastic signals using a set of statistical models and state transition probabilities. HMM has been successfully applied for speech recognition where time alignment problem is series. Also, HMM was extended to face recognition task in recent years. In HMM model training phase, Viterbi algorithm is performed to predict a most likely state sequence to describe observation sequence. Training observations aligned by the corresponding states and used to estimate the state-based model parameters. Obviously, it is important to align training data by optimal states so as to estimate good model parameters. In this thesis, we develop a new objective criterion base on Maximum Confidence Measure for discriminative HMM models training. A discriminative transformation is incorporated into the objective criterion to extract low-dimensional and discriminative features. This transformation is able to improve training data alignment for HMM training. We demonstrate that this linear transformation is the solution of linear discriminant analysis under general assumptions. We exploit this new algorithm in the face recognition framework using embedded HMM (EHMM). The proposed algorithm is helpful for training robust and compact face models. Consequently, these models are able to describe the test face images with different variations.

　　Finally, we use ORL, FERET and UMIST facial databases to evaluate the performance of proposed algorithms. The recognition rate and computation time are compared in different cases. We find that new algorithm outperforms classical EHMM when suitable dimensional features are extracted. Segmentations of face images are also evaluated using different algorithms.

[1]M. Bicego, U. Castellani, V. Murino, “Using Hidden Markov Models and Wavelets for Face Recognition”, IEEE International Conference on Automatic Face and Gesture Recognition, 28-30, pp.257-262, March 2000
[2]P. Bickel, L. Doksum, Mathematical Statistics: Basic Idea and Select Topics. Englewood Cliffs, NJ: Prentice-Hall, 1997
[3]P. Bickel, K. Doksum, Mathematical Statistics: Basic Idea and Selected Topics. Englewood Cliffs, NJ: Prentice-Hall, 1977
[4]E. L. Bocchieri, J. G. Wilpon, “Discriminative Feature Selection for Speech Recognition”, Computer Speech and Language, Vol.7, pp.229-246, 1993
[5]R. Chellappa, C. L. Wilson, S. Sirohey, “Human and Machine Fecognition of Face: a Survey”, Proceedings of the IEEE, Vol.83, No.5 ,pp.705-741, May 1995
[6]W. Chou, B. H. Juang, C. H. Lee, “Segmental GPD Training of HMM-Based Speech Recognizer”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.1, pp.473-476, Apr. 1992
[7]G. Doddington, “Phonetically Sensitive Discriminants for Improved Speech Recognition”, IEEE International Conference on Acoustics Speech and Signal Processing, no.S10b.11, pp.556-559, 1989
[8]S. Eickeler, M. Jabs, G. Rigoll, “Comparison of Confidence Measure for Face Recognition”, IEEE International Conference on Automatic Face and Gesture Recognition, pp.257-262, March 2000
[9]K. Fukunaga, Introduction to Statistical Pattern Recognition, Academic Press, second edition, 1991
[10]D. B. Graham, N. M. Allinson, “Characterizing Virtual Eigensignatures for General Purpose Face Recognition”, Computer and Systems Sciences, Vol.163, pp.446-456, 1998
[11]X. D. Huang, Y. Ariki, M. A. Jack, Hidden Markov Models for Speech Recognition, Edinburgh University Press, 1990
[12]B.-H. Juang, S. Katagiri, “Discriminative Learning for Minimum Error Classification”, IEEE Transactions on Signal Processing, Vol.40, No.12, pp.3043-3054, Dec 1992
[13]V. V. Kohir, U. B. Desai, “Face Recognition Using A DCT-HMM Approach”, IEEE Workshop on Applications of Computer Vision, pp.226-231, Oct 1998
[14]M. S. Kim, D. Kim, S. Y. Lee, “Face Recognition Using the Embedded HMM with Second-Order Block-Specific Observations”, Pattern Recognition, Vol.36, No.11, pp.2723-2735, March 2003
[15]S. S. Kuo, O. E. Agazzi, “Keyword Spotting in Poorly Printed Documents Using Pseudo 2-D Hidden Harkov Model”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.16, No.8, pp.842-848, Aug 1994
[16]N. Kumar, A. G. Andreou, “Heteroscedastic Discriminant Analysis and Reduced Rand HMM for Improved Speech Recognition”, Speech Communication, Vol.26, No.4, pp.283-297, Dec 1998
[17]N. Kumar, Investigation of Silicon Auditory Models and Generalization of Linear Discriminant Analysis for Improved Speech Recognition. Ph.D. thesis, John Hopkins University, 1997
[18]Hong-Kwang J. Kuo, Chin-Hwi Lee, I. Zitouni, E. Fosler-Lussier, “Minimum Verification Error Training for Topic Verification”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.1, 6-10, pp.1-380 – 1-383, April 2003
[19]D. Little, S. Krishna, J. Black, S. Panchanathan, “A Methodology for Evaluating Robustness of Face Recognition Algorithm with Respect to Variations in Pose Angle and Illumination Angle”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.6, 5-9, pp.89-92, March 2005
[20]Gu Liang, K. Rose, “Sub-State Tying in Tied Mixture Hidden Markov Model”, IEEE International Conference on Acoustics Speech and Signal Processing, vol.2, 5-9, pp.II1013-II1016, June 2000
[21]Chih-Pin Liao; Jen-Tzung Chien, “Nonsingular Discriminant Feature Extraction for Face Recognition”, IEEE International Conference on Acoustics Speech and Signal Processing, vol.2, pp.957-960, March.2005
[22]A. M. Martinez, A. C. Kak, “PCA versus LDA”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.23, PP228-233, February 2001
[23]A. V. Nefian, M. H. Hayes III, “Face Detection and Recognition Using Hidden Markov Models”, IEEE International Conference on Image Processing, Vol.1, 4-7, pp.141-145, Oct 1998
[24]A. V. Nefian, M. H. Hayes III, “Maximum Likelihood Training of the Embedded HMM for Face Detection and Recognition”, IEEE International Conference on Image Processing, Vol.1, 10-13, pp.33-36, Spet. 2000
[25]A. V. Nefian, M. H. Hayes III, “Hidden Markov Models for Face Recognition”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.5, pp.2721-2724, May 1998
[26]A. V. Nefian, M. H. Hayes III, “An Embedded HMM-Based Approach for Face Detection and Recognition”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.6, 15-19, pp.3553-3556, March 1999
[27]H. Othman, T. Aboulnasr, “Low Complexity 2-D Hidden Markov Model for face Recognition”, IEEE International Symposium on Circuits and Systems, Vol.5, pp.33-36, May 2000
[28]H. Othman, T. Aboulnasr, “Hybrid Hidden Markov Model for Face Recognition”, IEEE Southwest Symposium Image Analysis and Interpretation, 2-4, pp.36-40, April 2000
[29]H. Othman, T. Aboulnasr, “A Separable Low Complexity 2D HMM with Application to Face Recognition”, IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol.25, No.10, pp.1229-1238, Oct 2003
[30]H. Othman, T. Aboulnasr, “A Tied-Mixture 2D HMM Face Recognition System”, IEEE International Conference on Pattern Recognition, Vol.2, 11-15, pp.453-456, Aug.2002
[31]P. J. Phillips, H. Moon, P. J. Rauss, S. Rizvi, “The FERET Evaluation Methodology for Face Recognition Algorithms”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.22, No.10, Oct.2000
[32]H. S. Park, S. W. Lee, “A Truly 2D Hidden Markov Model for Off-Line Handwritten Character Recognition”, Pattern Recognition, Vol.31, No.12, pp.1849-1864, Dec 1998
[33]L. R. Rabiner, “A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition”, Proceedings of the IEEE, Vol.77, No.2, pp.257-286, Feb. 1989
[34]A. E. Rosenberg, O. Siohan, S. Parthasarathy, “Speaker Verification Using Minimum Verification Error Training”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.1, 12-15, pp.105-108, May 1998
[35]L. R. Rabiner, B. H. Juang, Fundamentals of Speech Recognition. Englewood Cliffs, NJ: Prentice-Hall, 1993
[36]M. G. Rahim, Chin-Hui Lee, Biing-Hwang Juang, Wu Chou, “Discriminative Utterance Verification Using Minimum String Verification Error (MSVE) Training”, IEEE International Conference on Acoustics Speech and Signal Processing,Vol.6, 7-10, pp.3585-3588, May 1996
[37]A. R. Setlur, R. A. Sukkar, J. Jacob, “Correcting Recognition Errors via Discriminative Utterance Verification”, IEEE International Conference on Spoken Language Proceedings, Vol.2, 3-6, pp.602-605, Oct.1996
[38]M. G. Rahim, Chin-Hui Lee, Biing-Huang Juang, “Discriminative Utterance Verification for Connected Digits Recognition”, IEEE Transactions on Speech and Audio Processing, Vol.5, No.3, pp.266-277, May 1997
[39]F. S. Samaria, S. Young, “HMM-based architecture for face identification”, Image and Vision Computing, Vol.12, No.8, pp.537-543, Oct 1994
[40]F. S. Samaria, “Face Recognition Using Hidden Markov Models”, Ph.D. Thesis, University of Cambridge, UK, 1994
[41]Rafid A. Sukkar, “Subword-Based Minimum Verification Error (SB-MVE) Training for Task Independent Utterance Verification”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.1, 12-15, pp.229-232, May 1998
[42]R. A. Sukkar, A. R. Setlur, M. G. Rahim, Chin-Hui Lee, “Utterance Verification of Keyword Strings Using Word-Base Minimum Verification Error (WB-MVE) Training”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.1, 7-10, pp.518-521, May 1996
[43]F. Wallhoff, S. Eickeler, G. Rigoll, “A Comparison of Discrete and Continuous Output Modeling Techniques for a Pseudo-2D Hidden Markov Model Face Recognition System”, IEEE International Conference on Image Processing, Vol.2, 7-10, pp.685-688, Oct 2001
[44]P. C. Woodland, D. R. Cole, “Optimising Hidden Markov Models Using Discriminative Output Distribution”, IEEE International Conference on Image Processing, Vol.1, pp.545-548, 1991
[45]F. Wessel, H. Ney, “Unsupervised Training of Acoustic Models for Large Vocabulary Continuous Speech Recognition”, IEEE Workshop on Automatic Speech Recognition and Understanding, 9-13, pp.307-310, Dec.2001
[46]Y. Q. Xu, B. C. Li, B. Wang, “Face Detection and Recognition Using Neural Network and Hidden Markov Models”, IEEE International Conference on Neural Networks and Signal Processing, Vol.1, 14-17, pp.228-231, Dec 2003
[47]Jian Yang, and Jing-yu Yang, “Why Can LDA be Performed in PCA Transformed Space?”, Pattern Recognition, Vol.36, pp.563-566, Feb 2003
[48]S. A. Zahorian, D. Qian, A. J. Jagharghi, “Acoustic-phonetic Transformations for Improved Speaker-Independent Isolated Word Recognition”, IEEE International Conference on Acoustics Speech and Signal Processing, Vol.1, pp.561-564, 1991
[49]吳佳珍, “以鑑別性小波參數為主之人臉辨識研究系統”, 國立成功大學資訊工程學系碩士論文, June 2001