Date of Award
2023
Publication Type
Dissertation
Degree Name
Ph.D.
Department
Electrical and Computer Engineering
Keywords
Face recognition, Feature extraction, Hierarchical, Preprocessing, Verification
Supervisor
M.Ahmadi
Supervisor
R.Riahl
Rights
info:eu-repo/semantics/openAccess
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Abstract
Human face recognition has become one of the most attractive topics in the fields of biometrics due to its wide applications. The face is a part of the body that carries the most information regarding identification in human interactions. Features such as the composition of facial components, skin tone, face's central axis, distances between eyes, and many more, alongside the other biometrics, are used unconsciously by the brain to distinguish a person. Indeed, analyzing the facial features could be the first method humans use to identify a person in their lives.
As one of the main biometric measures, human face recognition has been utilized in various commercial applications over the past two decades. From banking to smart advertisement and from border security to mobile applications. These are a few examples that show us how far these methods have come. We can confidently say that the techniques for face recognition have reached an acceptable level of accuracy to be implemented in some real-life applications. However, there are other applications that could benefit from improvement. Given the increasing demand for the topic and the fact that nowadays, we have almost all the infrastructure that we might need for our application, make face recognition an appealing topic.
When we are evaluating the quality of a face recognition method, there are some benchmarks that we should consider: accuracy, speed, and complexity are the main parameters. Of course, we can measure other aspects of the algorithm, such as size, precision, cost, etc. But eventually, every one of those parameters will contribute to improving one or some of these three concepts of the method. Then again, although we can see a significant level of accuracy in existing algorithms, there is still much room for improvement in speed and complexity. In addition, the accuracy of the mentioned methods highly depends on the properties of the face images. In other words, uncontrolled situations and variables like head pose, occlusion, lighting, image noise, etc., can affect the results dramatically.
Human face recognition systems are used in either identification or verification. In verification, the system's main goal is to check if an input belongs to a pre-determined tag or a person's ID.
Almost every face recognition system consists of four major steps. These steps are pre-processing, face detection, feature extraction, and classification. Improvement in each of these steps will lead to the overall enhancement of the system. In this work, the main objective is to propose new, improved and enhanced methods in each of those mentioned steps, evaluate the results by comparing them with other existing techniques and investigate the outcome of the proposed system.
Recommended Citation
Parsai, Soroosh, "Improved Human Face Recognition by Introducing a New Cnn Arrangement and Hierarchical Method" (2023). Electronic Theses and Dissertations. 9045.
https://scholar.uwindsor.ca/etd/9045