A statistically rigorous approach to the aging of the human face

The ability to accurately age an image of the human face in an automatic and rigorous fashion has widespread potential applications. This thesis is concerned with the development and testing of a new approach to computerised age progression based on a statistical learning procedure. The thesis begins with an overview of existing methodologies for age progression and outlines the need for improved procedures. After a review of the underpinning mathematical techniques, the theoretical basis of the new age-progression methodology is then presented. In this new approach, age progression is achieved through the calculation of optimised trajectories within a model space constructed from a principal component analysis of the shape and texture of a training sample of images. The statistical framework proposed extends naturally to include both generic and person-specific influences on the changes in facial appearance as aging progresses. Specific, physiological developmental periods, facial appearance at a previous age and the tendency to resemble close relatives are all incorporated into the model. The methodology is then computationally implemented and tested. Quantitative and perceptual tests both confirm the essential validity and accuracy of the techniques. This new methodology demonstrates that near photographic-quality, age-progressed images may be obtained based on rigorous scientific principles and considerably more quickly than is currently possible using forensic artistry. It is concluded that the algorithms may, in the future, be used to augment or even replace the existing artistic methodology.