在這一篇論文中,我們提出了一個克服不同光源變化的新人臉辨識演算法。我們的方法主要是以Fisherface分析為基礎並包含了三個有效的策略去克服光源變化的問題。第一個策略是使用了影像的梯度特徵而非亮度特徵，這樣的特徵比較不受光源變化的影響。第二個策略是影像的分割並且使用可調適的區塊比對。我們適當的將人臉區域切成數個方塊來做比對，並且根據PCA分析方法重建回來的誤差給予比較可靠的區域一個比較大的比重值。藉由此方式，資料遺失的區域將被給予一個較小的比重值，藉此我們可以降低資料遺失區塊做辨識時所造成的影響。第三個策略則是我們在兩張人臉影像間使用了新的相似度計算方法。此相似度計算方式是以L2 and Mahalanobis-L2 distances為基礎並結合PCA跟LDA的係數去決定相似度。藉由引入這一些策略到我們提出的演算法中，我們能夠達到更強固的人臉辨識。
經由在不同人臉資料庫的實驗，我們證明了我們提出的演算法明顯的比其他知名的方法更為精確，而且值得注意的是當光源變化造成大的陰影情況下，我們提出的演算法比其他方法有更明顯的優勢。另外，當每個人只有一張訓練資料的情況下，我們提出的分割性的梯度PCA演算法依然可以證明是一個成功的辨識方法。我們的實驗結果顯示出我們的演算法成功的克服了人臉辨識上光源變化的問題並且在實際應用上是可行的。

In this thesis, we propose a novel algorithm for face recognition under illumination variations. Our approach is mainly based on the Fisherface analysis and consists of three effective strategies to overcome the illumination variation problem. The first one is the use of the image gradient features instead of intensity features to be robust against illumination variations. The second one is image partition with adaptive block comparison. We appropriately divide the face region into several blocks for comparison and assign large weights to reliable blocks depending on the relative PCA reconstruction error. By the way, the regions of missing data will be given a smaller weight and then we can reduce the effect of missing data. The third one is the use of a novel similarity measure between face images. This measure is based on the L2 and Mahalanobis-L2 distances and combines both PCA and LDA coefficients to determine the similarity. By employing these strategies into the proposed algorithm, we can achieve more robust recognition.
We show that the proposed algorithm significantly outperforms other well-known appearance-based approaches through experiments on several benchmarking face databases. The superior performance if the proposed algorithm becomes notable for severe lighting conditions that introduce strong shadows. The proposed partitioned gradient PCA algorithm is proved to be successful in the face recognition experiments when only one training image per person is available. Our experimental results show that our algorithm successfully overcomes the illumination variation problem in face recognition and it is feasible for practical applications.

[1] P.N. Belhumeour, J.P. Hespanpa, and D.J. Kriegman, “Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 711-720, 1997.
[2] L Chengiun, “A Bayesian Discriminating Features Method for Face Detection,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 25, no. 6, pp. 725-740, 2003.
[3] C.P. Papageorgiou, M. Oren, and T. Poggio, “A General Frame Work for Object Detection,” Proc. Int. Conf. on Computer Vision, pp.555-562, Jan 1998.
[4] M. Turk, and A. Penland, “Eigenfaces for Recognition,” J. Cognitive Neuroscience, vol. 3, no. 1, pp. 71-86, 1991.
[5] J. Ross Beveridge. “The Geometry of LDA and PCA Classifiers Illustrated with 3D examples,” Technical Report CS-01-101, Computer Science, Colorado State University, 2001.
[6] W. Zhao, R. Chellappa, and P.J. Phillips, “Subspace Linear Discriminant Analysis for Face Recognition,” in UMD, 1999.
[7] A.M. Martinez, and A.C. Kak, “PCA versus LDA,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 23, no. 2, pp. 228-233, 2001.
[8] R. Duda, and P. Hart, Pattern Classification and Scene Analysis. New York: Wiley, 1973.
[9] H. Yu, and J. Yang, “A Direct LDA Algorithm for High-Dimensional Data - With Application to Face Recognition,” Pattern Recognition, vol. 34, no. 10, pp.2067-2070, 2001.
[10] W. Zhao, R. Chellappa, A. Rosenfild, and P.J. Phillips, “Face Recognition: A Literature Survey,” University of Maryland, CS-Tech Report-4167, 2000.
[11] R. Chellappa, C.L. Wilson, and S. Sirohey, “Human and Machine Recognition of Faces: A Survey,” Proc. IEEE, vol. 83, pp. 705-740, 1995.
[12] M.D. Kelly, “Visual Identification of People by Computer,” Tech. Rep. AI-130, Stanford SI Proj., Stanford CA, 1970.
[13] T. Kanade, Computer Recognition of Human Faces. Basel and Stuttgart: Birkhauser, 1977.
[14] R. Brunelli, and T. Poggio, “Face recognition: Features versus Templates,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 15, no. 10, pp. 1042-1053, 1993.
[15] B. Scholköpf, A. Smola, and K.R. Müller, “Nonlinear Component Analysis as A Kernel Eigenvalue Problem,” Neural Computation, vol. 10, no. 5, pp. 1299-1319, 1998.
[16] M.H. Yang, N. Ahuja, and D. Kriegman, “Face Recognition Using Kernel Eigenfaces,” in Proc. Int. Conf. Image Processing, pp. 37-40. 2000.
[17] Q.S. Liu, R. Huang, H.Q. Lu, and S.D. Ma, “Face Recognition Using Kernel Based Fisher Discriminant Analysis,” in Proc. Int. Conf. Automatic Face and Gesture Recognition, Washington, DC, pp. 197-201, 2002.
[18] Q.S. Liu, H.Q. Lu, and S.D. Ma, “Improving Kernel Fisher Discriminant Analysis for Face Recognition,” IEEE Trans. Circuits and System for Video Technology, vol. 14, no. 1, pp. 42-49, 2004.
[19] B. Heisele, P. Ho, and T. Poggio, “Face Recognition with Support Vector Machines: Global versus Component-based Approach,” IEEE Proc. Int. Conf. on Computer Vision, vol. 2, pp. 7-14, 2001.
[20] G. Gordon, “Face Recognition Based on Depth Maps and Surface Curvature,” in SPIE Proceedings, vol. 1570: Geometric Method in Computer Vision, pp. 234-247, 1991.
[21] L. Sirovich, and M. Kirby, “Low-Dimensional Procedure for Characterization of Human Faces,” J. Optical Society American, vol. 4, pp.519-524, 1987.
[22] M. Kirby, and L. Sirovich, “Application of the KL Procedure for the Characterization of Human Faces,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 12, no.1 pp. 103-108, 1990.