Higgins, Jill Eileen
Curve extraction and facial analysis using statistical techniques.
MSc(R) thesis, University of Glasgow.
Full text available as:
The analysis of shape is an area of statistics which has wide applications, and is made particularly interesting by its visual nature. The field of facial shape analysis is particularly interesting because the human face has such a complex but familiar shape. This thesis investigates the facial characteristics of a cohort of 5-year-old children, whose faces were photographed, and the corresponding data captured, to provide a control group with which to undertake a study into Cleft Lip or Cleft Lip and Palate.
Chapter 1 provides an introduction to the data, in particular to the defined landmarks of the face, and to the shape analysis theory necessary for much of the analysis.
Chapter 2 involves the analysis of the landmark-based data, including consideration of the reliability with which landmarks are identified. The issue of sexual dimorphism is then introduced, and we consider whether any significant differences exist between the males and females of the cohort, including the use of principal components analysis to explore the data.
To expand upon the landmark-based analysis, the third and fourth chapters involve the development of methods with which to extract a set of facial curves.
Chapter 3 uses a combination of 2 methods for defining planes in three dimensional space with which to "cut" the face, and then identify the curve which lies at the intersection of the face and the plane. The set of facial curves identified using this method are then analysed using the same principal components analysis that was used for the landmark-based data.
Chapter 4 introduces theory and methodology to identify the facial curves based on the surface curvature characteristics of the face, and goes on to explore interesting features which were observed as part of the curve extraction.
Chapter 5 discusses the findings of the thesis and provides some thoughts on further work which would be of interest.
Actions (login required)