Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Lane Department of Computer Science and Electrical Engineering

Committee Chair

Guodong Guo

Committee Member

Matthew C. Valenti

Committee Member

Xin Li

Committee Member

Donald Adjeroh

Committee Member

Hong-Jian Lai


In the past few decades, overweight and obesity are spreading widely like an epidemic. Generally, a person is considered overweight by body mass index (BMI). In addition to a body fat measurement, BMI is also a risk factor for many diseases, such as cardiovascular diseases, cancers and diabetes, etc. Therefore, BMI is important for personal health monitoring and medical research. Currently, BMI is measured in person with special devices. It is an urgent demand to explore conveniently preventive tools. This work investigates the feasibility of analyzing BMI from human visual appearances, including 2-dimensional (2D)/3-dimensional (3D) body and face data.

Motivated by health science studies which have shown that anthropometric measures, such as waist-hip ratio, waist circumference, etc., are indicators for obesity, we analyze body weight from frontal view human body images. A framework is developed for body weight analysis from body images, along with the computation methods of five anthropometric features for body weight characterization. Then, we study BMI estimation from the 3D data by measuring the correlation between the estimated body volume and BMIs, and develop an efficient BMI computation method which consists of body weight and height estimation from normally dressed people in 3D space.

We also intensively study BMI estimation from frontal view face images via two key aspects: facial representation extracting and BMI estimator learning. First, we investigate the visual BMI estimation problem from the aspect of the characteristics and performance of different facial representation extracting methods by three designed experiments. Then we study visual BMI estimation from facial images by a two-stage learning framework. BMI related facial features are learned in the first stage. To address the ambiguity of BMI labels, a label distribution based BMI estimator is proposed for the second stage. The experimental results show that this framework improves the performance step by step. Finally, to address the challenges caused by BMI data and labels, we integrate feature learning and estimator learning in one convolutional neural network (CNN). A label assignment matching scheme is proposed which successfully achieves an improvement in BMI estimation from face images.

Embargo Reason

Publication Pending