Facial expression recognision based on scale invariant feature transform
Çakır, Demet
Facial expression recognision based on scale invariant feature transform Demet Çakır; Supervisor: Mehmet Kuşaf - Nicosia Cyprus International University 2010 - X, 59 p ill, fig 30.5 cm
include references (55-59 p.)
CHAPTER 1-INTRODUCTION 1 Overview 1 Problem Statement and Its Significance 2 Literature Survey 4 Geometric Features 5 Apperance Features 6 Model Template-Based Approaches 9 Local Descriptors 11 Contribution to Facial Representaion 12 Feature Representation 12 Feature Extraction and Selection 13 Classifier Design 13 Thesis Overview 14 CHAPTER 2-SIFT METHODOLOGY 16 Scale-Space Extreme Detection 16 Keypoint Localization 19 Keypoint Descriptor 21 CHAPTER 3-FEATURE REPRESENTATION FOR MATCHING 24 Basic Concepts of Feature Representation 24 SIFT Feature Representation using Face Descriptors 24 Feature Matching 26 Face Descriptor Matching 27 Improved SIFT Matching Using Kullback-Leibler Divergence 29 Evaluation and Experiment Results 30 Conclusions 32 CHAPTER 4: FEATURE EXTRACTION AND SELECTION 33 Feature Selection 33 Feature Selection 34 Purpose and Importance of Feature Selection 34 Choosing among Features 35 Related Work 36 Discriminating Power Estimation of SIFT Features 37 BU-3DFE Database 38 Evaluation and Experiment Result 39 Conclusions 44 CHAPTER 5-CLASSIFIER DESIGN FOR SIFT FEATURES 45 Classifier Design 45 Image Matching using Regular SIFT 45 Minimum Pair Distance (MPD) 46 Sub-region Construction for Feature Matching 47 Regular Grid Base Face Descriptor Matching 48 Evaluation and Experimental Results 49 Conclusion 51 CHAPTER 6-CONCLUSION AND FUTURE WORKS 53 Conclusions 53 Future Works 53 REFERENCES 55
' Human computer interaction (HCL) is an active research area since the early days of computers. HCL aims to ease computer usage through various methods. One of the most important method is to act on human behavior. In this problem of person independent facial expression recognition from Scale Invariant Features Transform (SIFT) is investigated. Since human communication is mainly performed by face, we have chosen to extract facial expressions. Our primary objective is to increase available identification performance in the literature, while trying to solve problems related to external facts, such as lighting orientation and scaling. In our research, we propose a feature selection methodology that uses key point descriptors of the SIFT features features extracted from the expression images to improve classifications. The selection algorithm aims to derive a set of SIFT features from the original expression images which minimizes the within-class variance. The proposed methodology is evaluated using 3D facial expression data base BU-3DFE (Binghamton University 3D Facial Expression). Kullback Leibler Divergence (KLD) is used as a distance measure for computing the dissimilar score between histograms. Facial expression such as anger, fear, happiness, sadness and surprise are successfully recognized. '
Scale invariant feature transform
Ölçek değişmez özellik dönüşümü
Kullback Leibler Divergence
Kullback Leibler Divergence
Facial expression recognision based on scale invariant feature transform Demet Çakır; Supervisor: Mehmet Kuşaf - Nicosia Cyprus International University 2010 - X, 59 p ill, fig 30.5 cm
include references (55-59 p.)
CHAPTER 1-INTRODUCTION 1 Overview 1 Problem Statement and Its Significance 2 Literature Survey 4 Geometric Features 5 Apperance Features 6 Model Template-Based Approaches 9 Local Descriptors 11 Contribution to Facial Representaion 12 Feature Representation 12 Feature Extraction and Selection 13 Classifier Design 13 Thesis Overview 14 CHAPTER 2-SIFT METHODOLOGY 16 Scale-Space Extreme Detection 16 Keypoint Localization 19 Keypoint Descriptor 21 CHAPTER 3-FEATURE REPRESENTATION FOR MATCHING 24 Basic Concepts of Feature Representation 24 SIFT Feature Representation using Face Descriptors 24 Feature Matching 26 Face Descriptor Matching 27 Improved SIFT Matching Using Kullback-Leibler Divergence 29 Evaluation and Experiment Results 30 Conclusions 32 CHAPTER 4: FEATURE EXTRACTION AND SELECTION 33 Feature Selection 33 Feature Selection 34 Purpose and Importance of Feature Selection 34 Choosing among Features 35 Related Work 36 Discriminating Power Estimation of SIFT Features 37 BU-3DFE Database 38 Evaluation and Experiment Result 39 Conclusions 44 CHAPTER 5-CLASSIFIER DESIGN FOR SIFT FEATURES 45 Classifier Design 45 Image Matching using Regular SIFT 45 Minimum Pair Distance (MPD) 46 Sub-region Construction for Feature Matching 47 Regular Grid Base Face Descriptor Matching 48 Evaluation and Experimental Results 49 Conclusion 51 CHAPTER 6-CONCLUSION AND FUTURE WORKS 53 Conclusions 53 Future Works 53 REFERENCES 55
' Human computer interaction (HCL) is an active research area since the early days of computers. HCL aims to ease computer usage through various methods. One of the most important method is to act on human behavior. In this problem of person independent facial expression recognition from Scale Invariant Features Transform (SIFT) is investigated. Since human communication is mainly performed by face, we have chosen to extract facial expressions. Our primary objective is to increase available identification performance in the literature, while trying to solve problems related to external facts, such as lighting orientation and scaling. In our research, we propose a feature selection methodology that uses key point descriptors of the SIFT features features extracted from the expression images to improve classifications. The selection algorithm aims to derive a set of SIFT features from the original expression images which minimizes the within-class variance. The proposed methodology is evaluated using 3D facial expression data base BU-3DFE (Binghamton University 3D Facial Expression). Kullback Leibler Divergence (KLD) is used as a distance measure for computing the dissimilar score between histograms. Facial expression such as anger, fear, happiness, sadness and surprise are successfully recognized. '
Scale invariant feature transform
Ölçek değişmez özellik dönüşümü
Kullback Leibler Divergence
Kullback Leibler Divergence