상세정보

The acquisition and analysis of facial expression information is the key to achieving natural communication in human computer interaction. Therefore, research on facial expression recognition has attracted considerable attention in recent years. Generally speaking, the process of facial expression recognition includes image preprocessing, face detection, facial feature location, expression feature extraction, and expression classification. Based on the existing studies, this dissertation has done some research on the algorithms in the components of facial expression recognition. The main research contents are as follows.
(1) Research on image preprocessing. According to the modularity principle of visual information processing, a fractional step color constancy algorithm for color images under complex illumination conditions is proposed. This algorithm can effectively eliminate color deviation and uneven illumination so that human face can be correctly detected.
(2) Research on face detection. Face detection is the basis of facial expression recognition. An improved AdaBoost face detection algorithm combined with skin color features is put forward. This algorithm makes it possible to detect human face accurately and efficiently.
(3) Research on facial feature location. A simple but efficient location method for key facial features which makes use of the characteristics of human face color information is presented. This method has high accuracy rate, and it is much robust for the variation of the expressions and poses.
(4) Research on facial expression image normalization. The geometrical characteristics and optical properties of facial expression image should be normalized before expression feature extraction. The pure expression image after normalization has uniform size, angle, and luminance, and removes the impact of the light and intensity of illumination.
(5) Research on expression feature extraction and expression classification. A facial expression recognition algorithm based on supervised orthogonal locality preserving projection is proposed. According to the characteristics of facial expressions, this algorithm firstly combines Gabor local statistical features with LBP texture features as composite expression features, and then reduces the feature dimension by applying supervised orthogonal locality preserving projection. Finally, facial expressions can be classified by adopting nearest neighbor algorithm. The proposed algorithm makes use of category prior knowledge to further improve classification performance.

ABSTRACT	i
ACKNOWLEDGEMENTS	iii
1. Introduction	1
1.1 Motivation and Purpose of the Research	1
1.2 Main Contents of the Research	2
1.3 Organization of the Dissertation	5
2. Background and Related Works	7
2.1 Image Preprocessing	7
2.1.1 Correction of Color Deviation	7
2.1.2 Luminance Enhancement	9
2.2 Face Detection	13
2.3 Facial Feature Location	14
2.4 Facial Expression Image Normaliztion	14
2.5 Facial Expression Feature Extraction	15
2.6 Facial Expression Recognition	16
3. Facial Expression Image Processing	18
3.1 Color Constancy Algorithm for Color Images under Complex Illumination Conditions	18
3.1.1 Automatic White Balance Algorithm Based on LoG Edges	18
3.1.2 Color Retention Luminance Enhancement Algorithm	27
3.1.3 Fractional Step Color Constancy Algorithm	34
3.2 Face Detection Algorithm Combining Skin Color Features with AdaBoost	37
3.2.1 Skin Color Clusting Analysis	37
3.2.2 Skin Color Area Detection Algorithm	41
3.2.3 Improved AdaBoost Face Detection Algorithm Combined with Skin Color Features	46
3.3 Key Facial Feature Location	49
3.3.1 Eye Location	49
3.3.2 Mouth Location	53
3.3.3 The Process of Key Facial Feature Location	57
3.4 Facial Expression Image Normalization	58
3.4.1 Rotation Normalization	58
3.4.2 Size Evaluation	60
3.4.3 Illumination Evaluation	62
4. Facial Expression Feature Extraction	64
4.1 Gabor Local Statistical Features	64
4.1.1 Gabor Filter	65
4.1.2 Expression Features Extracted from Gabor Local Statistical Features	66
4.2 Local Binary Pattern Features	68
4.2.1 Basic LBP and its improvement	69
4.2.2 Expression Features Extracted from an LBP Histogram	71
5. Facial Expression Recognition Based on SOLPP	74
5.1 Linear Discriminant Algorithm	74
5.2 Supervised Orthogonal Locality Preserving Projection	76
5.3 Facial Expression Classification	81
6. Experiment and Evaluation	83
6.1 Experiment to Evaluate Color Constancy Algorithm	83
6.1.1 Comparison of Automatic White Balance Algorithms	83
6.1.2 Comparison of Luminance Enhancement Algorithms	85
6.2 Experiment to Evaluate Face Detection Algorithm	88
6.3 Experiment to Evaluate Key Facial Feature Location	91
6.4 Experiment to Evaluate Expression Image Normalization	94
6.5 Experiment to Evaluate Expression Recognition Algorithm	96
7. Conclusion and Future Works	103
7.1 Conclusion	103
7.2 Future Works	104
Bibliography	106