Analysis and Segmentation of Facial Features in Laser Range Data*

This paper describes a new method for analysis and segmentation of facial laser range data employing complex–valued non–orthogonal Gabor wavelets, principal component analysis and a topological mapping network. The initial data set consisting of shape and texture information is encoded in amplitude and phase of a complex 2D image function. A set of oriented Gabor filters performs a decomposition of the data and allows for retrieving local shape and texture features. In order to cluster the obtained feature vector further subspace mapping methods have to be figured out. For this purpose, a segmentation pipeline is proposed consisting of principal component analysis, normalization and a topological mapping network. This process ultimately renders a R,G,B subspace representation of the multidimensional feature vector.