Title: Tensor-Factorisation-Based Modelling and Analysis of Human Faces

Wednesday 10 May 2017 at 1100-1200 in ITU 3A08

Abstract: Modelling and analysis of human faces has been a popular topic in computer vision, computer graphics, and computer games during last decades. This talk will present our recent research where we have proposed a tensor-factorisation-based computational model for human faces that naturally divides the faces into subspaces corresponding to shape, identity, and expression. Our approach is built upon the high-order singular value decomposition and it allows controlled estimation, morphing, transfer, and classification of facial shapes and expressions. As an interesting insight, we will also show that the expression subspace is star shaped. This stems from the observation that the trajectory obtained by increasing the strength of an expression approximately has a linear form in the expression subspace, and all these linear trajectories meet in a single point which can be seen as the root of all expressions. We will demonstrate modelling and analysis results by several experiments with real facial data. In contrast to the previous state-of-the-art, we will show, for instance, how we can morph and transfer faces and expressions between different people, without unwanted strong deformations, even if the people are not included in the training database. In the application of expression classification, the results are also promising.

Bio: Sami Brandt is PhD 2002 from Helsinki University of Technology and is associate professor at DIKU, University of Copenhagen, and Senior Mathematical Software Developer at 3Shape A/S.