Face Recognition using Local Feature Descriptors and Convolutional Neural Networks

Abstract

Face recognition is a process of verifying an individual using face images and it is widely employed in identifying people on social media platforms, validating identity at ATMs, finding missing persons, controlling access to sensitive areas, finding lost pets, etc. Face recognition, even though a well-known field of research over the decades, has still several open challenges. Few of these problems involve the recognition of faces with poor illumination, different poses, expressions, and occlusions. The traditional face recognition approaches can effectively recognize the faces, but the introduction of deep learning techniques has improved the recognition accuracy a lot in an unconstrained environment. newline With the aim of face recognition in an unconstrained environment, this research initially explored how the guided image filter (GIF) and a convolutional neural network (CNN) can enhance face recognition accuracy. Here, the Viola-Jones algorithm is used to find the face area and then smoothened by a GIF. Then, the proposed CNN is used to extract the features and recognize the faces. To support this work, a new face recognition method using a Non-Subsampled Shearlet Transform, histogram of Local Feature Descriptors, and a CNN is introduced. The local feature descriptors like pyramid of histogram of oriented gradients (PHOG), local phase quantization (LPQ), and the proposed CNN are used for extracting the features. The obtained features are fused to produce the feature vector and classified using support vector machine (SVM). newline Another approach for face recognition using CNN, histogram of local binary patterns (LBP), and histogram of oriented gradients (HOG) features is presented. The features from HOG, histogram of LBP, and deep features from the proposed CNN are linearly concatenated to produce the feature space and then classified by SVM. Finally, this thesis proposes a face recognition method using wavelet transform, center symmetric local binary patterns (CSLBP), and Dual-Stream CNN. The key