基于深度集成神经网络的人脸表情识别
发布时间：2024-02-24 21:58
近年来,深度学习方法极大地提高了人脸识别的准确性,为了获得更高的识别准确率,集成学习可以应用于深度学习算法中。传统识别算法难以捕捉到面部表情所传递的有用信息,面部表情识别存在分辨率低、遮挡、光照、位置等问题,通常情况下,由于这些面部表情分类很差,人类无法识别它们。此外,面部表情的分类比较特殊,例如面部微笑并不总是意味着开心,面部表情往往取决于文化。然而,提高面部表情识别准确率可以应用到更灵敏、更智能的系统,从而改善用户体验。为了提高分类器的性能,降低人脸表情识别的错误率,研究者开展了很多的工作,例如基于深度学习方法。有时候深度学习对面部表情识别存在困难,原因有很多,比如基于深度学习人脸面部表情识别应用是一项复杂而困难的任务,又例如很难找到高质量的数据集,深度网络的性能在很大程度上依赖于大量的标记样本。本文提出了一种基于卷积神经网络和集成深度网络的新方法,可面向小样本数据集分类情况,这些方法分别是多视角卷积神经网络(MVCNN)和集成迁移学习网络(ETLN)。首先,将人脸图像通过不同尺度进行下采样,然后向上采样到统一图像大小,得到多视角训练样本。然后,构造了一个具有双通道特征提取结构的多...
【文章页数】：85 页
【学位级别】：硕士
【文章目录】：
ABSTRACT
摘要
List of Symbols
List of Abbreviations
Chapter 1 Introduction
1.1 Background
1.2 Motivation of Our Work
1.3 Structure of The Thesis
Chapter 2 Related Works
2.1 Facial Expression Recognition
2.2 Literature Review
2.3 Neuron Model
2.4 Summary
Chapter 3 Multi-view Network based on CNN
3.1 Convolutional Neural Networks (CNN)
3.2 Multi-view CNN
3.2.1 Multiple View Datasets
3.2.2 Convolutional Layer
3.2.3 Pooling Layer
3.2.4 Fully Connected Layer
3.2.5 Batch Normalization Layer
3.2.6 Softmax Layer
3.2.7 Pre-Processing
3.2.8 Network Training
3.3 Datasets
3.3.1 The FER2013 Dataset
3.3.2 The RAF-BASIC Dataset
3.4 Results on FER2013 and Discussions
3.4.1 Experimental Condition
3.4.2 Results of DCNN with no data Aug
3.4.3 Results of DCNN with data Aug
3.4.4 Results of Multi-view CNN
3.5 Results on RAF-BASIC and Discussions
3.5.1 Results of DCNN with data Aug
3.5.2 Results of Transfer DCNN
3.6 Performance Evaluation of MVCNN and Transfer DCNN
3.7 Summary
Chapter 4 Ensemble Transfer Learning Network (ETLN)
4.1 Feature Learning
4.1.1 VGG16
4.1.2 VGG-face
4.1.3 Ensemble and Transfer Learning
4.1.4 Pre-Processing and Training Process
4.2 Experimental Details
4.2.1 Experimental Results on FER2013
4.2.2 Experimental Results on RAF-BASIC
4.3 Weights Analysis
4.4 Special Combination
4.5 Evaluation of The Proposed ETLN
4.6 Summary
Chapter 5 Conclusions and Future Work
5.1 Conclusions
5.2 Future Work
References
Acknowledgements
Biography
本文编号：3909618
【文章页数】：85 页
【学位级别】：硕士
【文章目录】：
ABSTRACT
摘要
List of Symbols
List of Abbreviations
Chapter 1 Introduction
1.1 Background
1.2 Motivation of Our Work
1.3 Structure of The Thesis
Chapter 2 Related Works
2.1 Facial Expression Recognition
2.2 Literature Review
2.3 Neuron Model
2.4 Summary
Chapter 3 Multi-view Network based on CNN
3.1 Convolutional Neural Networks (CNN)
3.2 Multi-view CNN
3.2.1 Multiple View Datasets
3.2.2 Convolutional Layer
3.2.3 Pooling Layer
3.2.4 Fully Connected Layer
3.2.5 Batch Normalization Layer
3.2.6 Softmax Layer
3.2.7 Pre-Processing
3.2.8 Network Training
3.3 Datasets
3.3.1 The FER2013 Dataset
3.3.2 The RAF-BASIC Dataset
3.4 Results on FER2013 and Discussions
3.4.1 Experimental Condition
3.4.2 Results of DCNN with no data Aug
3.4.3 Results of DCNN with data Aug
3.4.4 Results of Multi-view CNN
3.5 Results on RAF-BASIC and Discussions
3.5.1 Results of DCNN with data Aug
3.5.2 Results of Transfer DCNN
3.6 Performance Evaluation of MVCNN and Transfer DCNN
3.7 Summary
Chapter 4 Ensemble Transfer Learning Network (ETLN)
4.1 Feature Learning
4.1.1 VGG16
4.1.2 VGG-face
4.1.3 Ensemble and Transfer Learning
4.1.4 Pre-Processing and Training Process
4.2 Experimental Details
4.2.1 Experimental Results on FER2013
4.2.2 Experimental Results on RAF-BASIC
4.3 Weights Analysis
4.4 Special Combination
4.5 Evaluation of The Proposed ETLN
4.6 Summary
Chapter 5 Conclusions and Future Work
5.1 Conclusions
5.2 Future Work
References
Acknowledgements
Biography
本文编号：3909618
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/3909618.html