Lingyun Wen

Date of Graduation

2015

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Guodong Guo

Committee Co-Chair

Xin Li

Committee Member

Donald Adjeroh

Committee Member

Guodong Guo

Committee Member

Xin Li

Committee Member

Arun Ross

Committee Member

CunQuan Zhang

Abstract

The principal aim of facial image analysis in computer vision is to extract valuable information(e.g., age, gender, ethnicity, and identity) by interpreting perceived electronic signals from face images. In this dissertation, we develop facial image analysis systems for body mass index (BMI) prediction, makeup detection, as well as facial identity with makeup changes and BMI variations.;BMI is a commonly used measure of body fatness. In the first part of this thesis, we study BMI related topics. At first, we develop a computational method to predict BMI information from face images automatically. We formulate the BMI prediction from facial features as a machine vision problem. Three regression methods, including least square estimation, Gaussian processes for regression, and support vector regression are employed to predict the BMI value. Our preliminary results show that it is feasible to develop a computational system for BMI prediction from face images. Secondly, we address the influence of BMI changes on face identity. Both synthesized and real face images are assembled as the databases to facilitate our study. Empirically, we found that large BMI alterations can significantly reduce the matching accuracy of the face recognition system. Then we study if the influence of BMI changes can be reduced to improve the face recognition performance. The partial least squares (PLS) method is applied for this purpose. Experimental results show the feasibility to develop algorithms to address the influence of facial adiposity variations on face recognition, caused by BMI changes.;Makeup can affect facial appearance obviously. In the second part of this thesis, we deal with makeup influence on face identity. It is principal to perform makeup detection at first to address makeup influence. Four categories of features are proposed to characterize facial makeup cues in our study, including skin color tone, skin smoothness, texture, and highlight. A patch selection scheme and discriminative mapping are presented to enhance the performance of makeup detection. Secondly, we study dual attributes from makeup and non-makeup faces separately to reflect facial appearance changes caused by makeup in a semantic level. Cross-makeup attribute classification and accuracy change analysis is operated to divide dual attributes into four categories according to different makeup effects. To develop a face recognition system that is robust to facial makeup, PLS method is proposed on features extracted from local patches. We also propose a dual-attributes based method for face verification. Shared dual attributes can be used to measure facial similarity, rather than a direct matching with low-level features. Experimental results demonstrate the feasibility to eliminate the influence of makeup on face recognition.;In summary, contributions of this dissertation center in developing facial image analysis systems to deal with newly emerged topics effectively, i.e., BMI prediction, makeup detection, and the rcognition of face identity with makeup and BMI changes. In particular,to the best of our knowledge, BMI related topics, i.e., BMI prediction; the influence of BMI changes on face recognition; and face recognition robust to BMI changes are first explorations to the biometrics society.

Download

Share

COinS