Date of Award

2-7-2023

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Mariofanna Milanova

Abstract

Facial recognition refers to the determination of identity of an individual based on facial features. The application of facial recognition has been widely used for the purpose of security. The outbreak of coronavirus has made people wear masks as a preventive measure. However, this affects the proper identification of individuals resulting in several security threats. In this paper, the issues related to recognition of both masked and unmasked faces are mitigated and an approach for accurate detection of face masks and recognition of both masked and unmasked faces is Introduced. The proposed approach comprises several processes which are carried out in a step-by-step manner to achieve the objective. The CCTV surveillance camera videos are utilized as the input for this approach. The conversion of videos into frames is performed to execute the detection process. Initially, the selection of keyframes from the vast number of frames is carried out by using Histogram of Gradient (HoG) technique in order to focus mainly on those frames for the purpose of recognition. The augmentation of data is performed in order to contribute to the increased recognition rate which includes three processes such as normalization of color, correction of illumination, and normalization of poses. In color normalization, the RGB color channels are normalized into a single-color channel. The illumination correction is performed by utilizing parameterized Contrast Limited Adaptive Histogram Equalization (CLAHE) and the normalization of poses is implemented by using Angular Affine Transformation. These processes are performed in order to mitigate the limitations caused by variation in light in both indoor and outdoor environments. The augmented frames are then segmented by implementing Expectation-Maximization based Gaussian Mixture Model (EM-GMM) in which the face region is segmented from the other regions in order to focus mainly on the face regions. This greatly reduces the computation of the upcoming processes to be applied only in the segmented region. The facial features considered for effective segmentation of face region are forehead, chin, vii mouth, nose, and eyes. Once the segmentation is over, the extraction of pixel-based features for detection of face masks and recognition of masked and unmasked faces is carried out. The Yolo Nano model which is a lightweight and highly accurate model is utilized for this purpose. Both the high-level and low-level features are considered in order to perform precise detection of face masks. The construction of the Bag of Visual Words (BoVW) is carried out by clustering the features based on similarity. The Hassanat similarity is utilized for the purpose of similarity computation and the detection of face masks is carried out by implementing L2 distance function. The recognition of both masked and unmasked faces is performed by executing kernel-based Extreme Learning Machine (ELM) in which the selection of kernel function is carried out by using Slime Mould Optimization (SMO). The increased learning rate possessed by this approach contributes to increased accuracy in recognition of faces. The experimentation of the proposed approach is carried out in Python IDLE 3.8 tool by utilizing several image processing libraries. The validation of the proposed Yolo Nano approach is executed by comparing it with several existing approaches with respect to significant performance metrics such as segmentation accuracy, classification accuracy, precision, recall, F-measure, ROC curve, computation time, and confusion matrix. The numerical analysis is computed from which the efficiency of the proposed approach in detection of face masks and recognition of both masked and unmasked faces is determined. Keywords: COVID-19 pandemic, Face Mask Detection, Facial Recognition, CCTV surveillance camera, Keyframe selection, Data augmentation, Yolo Nano, KNN, and ArcFace.

Download

Share

COinS