Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.

Non-WSU users: Please talk to your librarian about requesting this thesis through interlibrary loan.

Access Type

WSU Access

Date of Award

January 2017

Degree Type

Thesis

Degree Name

M.S.

Department

Computer Science

First Advisor

Xuewen Chen

Abstract

In this dissertation, we present our contributions in face detection and facial race classification.

Face detection in unconstrained images is a traditional problem in computer vision community. Challenges still remain. In particular, the detection of partially occluded faces with pose variations has not been well addressed. In the first part of this dissertation, our contributions are three-fold. First, we introduce our four image datasets consisting of large-scale labeled face dataset, noisy large-scale labeled non-face dataset, CrowdFaces dataset, and CrowdNonFaces dataset intended to be used for face detection training. Second, we improve Viola-Jones (VJ) face detection results by first training a Convolutional Neural Network (CNN) model on our noisy datasets. We show our improvement over the VJ face detector on AFW face detection benchmark dataset. However, existing partial occluded face detection methods require training several models, computing hand-crafted features, or both. Hence, we thirdly propose our Large-Scale Deep Learning (LSDL), a method that does not require training several CNN models or hand-crafted features computations to detect faces. Our LSDL face detector is trained on a single CNN model to detect unconstrained multi-view partially occluded and non-partially occluded faces. The model is trained with a large number of face training examples that cover most partial occlusions and non-partial occlusions facial appearances. The LSDL face detection method is achieved by selecting detection windows with the highest confidence scores using a threshold. Our evaluation results show that our LSDL method achieves the best performance on AFW dataset and a comparable performance on FDDB dataset among state-of-the-art face detection methods without manually extending or adjusting the square detection bounding boxes.

Many biometrics and security systems use facial information to obtain an individual identification and recognition. Classifying a race from a face image can provide a strong hint to search for facial identity and criminal identification. Current facial race classification methods are confined only to constrained non-partially occluded frontal faces. Challenges remain under unconstrained environments such as partial occlusions and pose variations, low illuminations, and small scales. In the second part of the dissertation, we propose a CNN model to classify facial races with partial occlusions and pose variations. The proposed model is trained using a broad and balanced racial distributed face image dataset. The model is trained on four major human races, Caucasian, Indian, Mongolian, and Negroid. Our model is evaluated against the state-of-the-art methods on a constrained face test dataset. Also, an evaluation of the proposed model and human performance is conducted and compared on our new unconstrained facial race benchmark (CIMN) dataset. Our results show that our model achieves 95.1% of race classification accuracy in the constrained environment. Furthermore, the model achieves a comparable accuracy of race classification compared to human performance on the current challenges in the unconstrained environment.

Off-campus Download

Share

COinS