與辦識有關的 AdaBoost 相關文獻中，大多是使用了 Haar-like 特徵，再加上AdaBoost 學習演算法以達到辨識的目的。因 Haar-like 特徵在訓練及比對過程中，需要大量的運算且辨識效果有限，故多輔以 AdaBoost 演算法，以達到可接受的辦識效果。
當物體有細微特徵的情況下，以 Haar-like 特徵的辨識效果較不理想。為提昇具細微特徵物體的辨識效果，本文提出利用 WHT (Walsh-Hadamard Transform) 取代Haar-like 特徵及以積分影像計算特徵的過程。
因 WHT 有豐富的基底，其中的低頻基底與 Haar-like 特徵極為相似。在以人臉辨識的實驗中證明，WHT 與 AdaBoost 演算法的組合，可與 Haar-like 特徵、積分影像和 AdaBoost 演算法的組合有相似的結果。
本論文最後獲得的結論為：WHT 與 AdaBoost 演算法的組合與 Haar-like 特徵與AdaBoost 演算法組合在人臉辨識的實驗中，可以有較佳的效果。當用於辨認非正面人臉時，也可以獲得較穩定的辨認率。在訓練樣本數的要求上，也較 Haar-like特徵與 AdaBoost 演算法組合來得低。

According to the publishes related to AdaBoost in face recognition, most of them are using Haar-like features and AdaBoost algorithm to achieve the purpose of recognition. The application of Haar-like features in the training and recognition processes, needs a lot of computation. The effect of recognition is also limited. Therefore, the AdaBoost algorithm is involved to achieve acceptable results.
In the case of object with small features, the application of Haar-like feature cannot obtain a good recognition result. To improve the results of recognizing object with small features, the WHT (Walsh-Hadamard Transform) is proposed to replace Haar-like feature and integral image in order to compute the feature values.
The low-frequency basis in WHT's rich basis are similar to Haar-like features. In the experiments of face recognition the combination of WHT and AdaBoost algorithm can obtain similar result the the combination of Haar-like features, integral image and AdaBoost algorithm.
The conclusions in this thesis are: the combination of WHT and AdaBoost algorithm is better than the combination of Haar-like features and
AdaBoost algorithm in the experiments of face recognition. When WHT is used to improve the recognition of non-frontal face, the more stable results are also obtained. The number of samples in the training process is also much lower than the combination of Haar-like features and AdaBoost algorithm.

誌謝 iii
摘要 iv
Abstract v
第1章 前言 1
1.1動機 1
1.2目的 1
1.3論文架構 2
第2章 相關研究 3
2.1介紹 3
2.2色彩特徵 3
2.2.1色彩空間 3
2.2.2灰階影像 5
2.3對比度 6
2.3.1直方圖 6
2.3.2對比度擴展 7
2.4AdaBoost演算法 8
2.4.1Haar-like特徵擷取 9
2.4.2積分影像 10
2.4.3AdaBoost學習演算法 12
2.5Walsh-Hadamard 轉換 13
2.5.1Hadamard 矩陣 14
2.5.2Hadamard 轉換 14
第3章 系統說明 17
3.1介紹 17
3.2 系統架構 18

[1] H.Y., Wu. and C.M.,Chang., “Studies of the Invisible Tags Based on Image Features,” Tatung University Department of Computer Science and Engineering Thesis for Master of Science, 2011.
[2] Lew, M.S., “Principles of Visual Information Retrieval,” USA, Springer, 2001.
[3] Paschos, G., “Perceptually uniform color spaces for color texture analysis: an empirical evaluation,” IEEE Transactions on Image Processing, pp. 932-937, June 2001.
[4] Hasan, Y.M.Y. and Karam, L.J., “Morphological Text Extraction From Image,” IEEE Transactions on Image Processing, November 2000.
[5] Hong, Zhang. and Jiulun, Fan.,“A Threshold Segmentation Method for Sparse Histogram Image ” Sixth International Conference on Fuzzy Systems and Knowledge Discovery, pp.340-344,2009.
[6] Huang, J. and Zabih, R., “Combining color and spatial information for content-based image retrieval,” European Conference on Digital Libraries, September 1998.
[7] Kearns, M.J. and Vazirani, U.V., “An Introduction to Computational Learning Theory,” The MIT Press, August 1994.
[8] T.C., Yang. and M.S., Wang,“Implementation of a Road Sign Recognition System Based on Integration of Adaboost Classifier and Support Vector Machine,”Thesis for Master of Science, National Cheng Kung University, 2009.
[9] Viola, P. and Jones, M.J., “Robust Real-Time Face Detetion,” International Journal of Computer Vision, pp.137-154, 2004.
[10] Georgiou, S. Koukouvinos, C. Seberry, J., "Hadamard matrices, orthogonal designs and construction algorithms,". Designs 2002: Further computational and constructive design theory. Boston: Kluwer. pp. 133–205.
[11] Townsend, W. J. and Thornton, M. A., “Walsh spectrum computations using Cayley graphs,” IEEE Midwest Symposium on Circuits and Systems,pp. 110-113,MWSCAS 2001.
[12] Rao, K., Devarajan, V., Vlasenko, V., and Narasimhan, M., "Cal-sal Walsh -Hadamard transform,"IEEE Transactions on Acoustics, Speech and Signal Processing,pp.605-607,1978.
[13] Lienhart, R. and Maydt, J., “An extended set of Haar-like features for rapid object detection,” International Conference on Image Processing, I-900-903 vol.1,2002.
[14] Yunyang, Y. Zhibo, G. and Jingyu.Y., “Multi-view Face Detection Based on the Enhanced AdaBoost Using Walsh Features,”Eighth ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing,pp.200-205,July 2007
[15] Zong-Ying Ho, Yung-Nien Sun,. "Computer Aided Flying Target Tracking Based on Predictive Dynamic Layer Representations,"Thesis for Master of Science,National Cheng Kung University,2006
[16] Ben-Artzi,G.Hagit Z. and Hel-Or,Y., “The Gray-Code Filter Kernels,” IEEE Transactions on Pattern Analysis and Machine Intelligence,pp.382- 393,March 2007
人臉訓練資料集
[17] BioID-Technology Research -BioID Face Database http://www.bioid.com/downloads/software/bioid-face-database.html
[18] (JAFFE) Database http://www.kasrl.org/jaffe.html
[19] CMU Face Detection Project
http://vasc.ri.cmu.edu//idb/html/face/frontal_images/index.html
[20] MIT-CBCL face recognition database
http://cbcl.mit.edu/software-datasets/heisele/facerecognition-database.html