洪一平臺灣大學：資訊工程學研究所江岳軒Chiang, Yueh-HsuanYueh-HsuanChiang2007-11-262018-07-052007-11-262018-07-052005http://ntur.lib.ntu.edu.tw//handle/246246/53885我們提出了一個新的適用於不同光源環境下的人臉辨識機制。在不同光源環境下的人臉影像是線性不可分的，而不同光源對人臉影像所帶來的變化遠比不同身份的人所帶來的還要大得多。此方法之基本概念是找幾組與光源環境相關的線性轉換向量，使得每一張影像將因為其光源的環境差異而使用不同程度的轉換、進而達成辨識的目標。所提出的機制可以分為幾個步驟，第一步驟是從光源的資料庫中找出一組最具代表性的光源環境類別(Lighting Condition Class)，其後我們將每組訓練影像用光源環境彈性分類的機制彈性分類到上面數個類別之中；最後，配合彈性分類的結果，我們將找出與光源環境相關的線性轉換向量來達成人臉辨識的目標。藉著彈性分類與線性轉換向量的特性，我們提出的方法除了可以避免過度配適(overfitting)的問題外、還同時擁有了低的計算量。採用我們提出的方法後，在不同光源環境下的人臉影像將可以良好地被區隔開。我們已經在若干個眾所周知的人臉資料庫測試了所提出的方法，實驗結果指出我們的方法比傳統的人臉辨識機制更好，並且更能表現人臉影像上光源的變化。We proposed a novel method of face recognition under varying lighting conditions. Face images under different lighting conditions are non-linear separable, image variation due to different lighting conditions is much more significant than that due to different personal identities. The basic idea of our approach is to find a set of lighting condition specific transformations which best separates the face images under varying lighting conditions. The proposed method has several steps, the first one is to find the optimal set of lighting condition classes which best describes the lighting variation, and then we apply a novel soft classification of lighting condition to each training image. With the soft classification result, a set of lighting condition specific linear transformations would be found to complete the recognition task. By the virtue of soft classification and linear transformations, our approach can not only avoid overfittings but also has low computational cost. With our method, face images under varying lighting conditions can be well separated. The proposed method has been tested on several well-known databases, and the experimental results show that the performance of our approach is better than those of conventional methods.List of Figures v List of Tables ix Chapter 1 Introduction 1 1.1 Overview 1 1.2 Conventional Approaches 1 1.2.1 Non-linear Approaches 2 1.2.2 Illumination Cones 3 1.2.3 Quotient Images and Self-Quotient Images 4 1.3 The Proposed Method 5 1.4 Organization 7 Chapter 2 Reviews of Discriminant Analysis 9 2.1 Linear Discriminant Analysis (LDA) 9 2.2 Generalized Discriminant Analysis (GDA) 10 2.3 Locally Linear Discriminant Analysis (LLDA) 12 Chapter 3 Selecting the Representative Lighting Condition Classes (LCCs) 15 3.1 Lighting Condition Class 16 3.1.1 Image Normalization 16 3.1.2 Combining LCCs Reflective to Each Other Horizontally 18 3.2 Lighting Correlation Image 19 3.3 Measures for Lighting Difference between Two Images 20 3.3.1 Euclidean Distance Metric 20 3.3.2 Dissimilarity based on Image Correlation 22 3.3.3 Dissimilarity based on Pixel-wise Correlation 24 3.4 Selecting the Representative Set of K LCCs 25 Chapter 4 Lighting Condition Class-based Locally Linear Discriminant Analysis (LCC+LLDA) 35 4.1 Soft Classification of Lighting Condition 37 4.1.1 Detecting Images Having Opposite Lighting Condition to the Representative LCCs 37 4.1.2 Soft Classification of Lighting Condition 38 4.2 Lighting Condition Class-Based LLDA 39 4.3 Gradient-Based Solution for LCC+LLDA 42 Chapter 5 Experiments 45 5.1 The Experimental Setup of Selecting the Representative Lighting Condition Classes 45 5.1.1 Database 45 5.1.2 Setup 49 5.2 Database 49 5.2.1 The CMU PIE Database 50 5.2.2 The BANCA Database 50 5.3 Experimental Results 52 5.3.1 Experiments based on the Yale Database B 52 5.3.2 Experiments based on the CMU PIE Database 53 5.3.3 Experiments based on the BANCA Database 53 5.3.4 Experiments under Different Number of Representative LCCs 55 5.4 Discussion 56 Chapter 6 Conclusions and Future Work 59 6.1 Conclusions 59 6.2 Future Work 60 Reference 637350045 bytesapplication/pdfen-US人臉辨識光源變化彈性分類局部線性鑑別分析face recognitionlighting variationsoft classificationlocally linear discriminant analysisthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53885/1/ntu-94-R92922029-1.pdf