The thesis presents the PISI model for face recognition. It represents an attempt at solving the problem of determining whether two given facial images represent the same subject, insensitive to the pose of the subject.
The model is based on deep convolutional neural networks. The fully-trained model is evaluated using the standard facial image data sets FERET and IJB-A using performance curves and quantitative performance metrics. The PISI model is found to outperform the PCA and LDA methods applied on Gabor features, on both data sets.
Chapter 1 provides an introduction to the thesis. Chapter 2 presents a historical overview of the fields of deep learning and facial biometry. Chapter 3 contains an explanation of the theory of the structure and learning process of deep neural networks. Chapter 4 describes the PISI model and the learning methods used to set its parameters. Chapter 5 presents the databases used to train and validate the model and methods used to evaluate the results, and chapter 6 presents the results of the evaluation. Chapter 7 presents the
|