Wedgelet Enhanced Appearance Model

Statistical region-based segmentation methods such as the Active Appearance Model (AAM) are used for establishing dense correspondences in images based on learning the variation in shape and pixel intensities in a training set. For low resolution 2D images this can be done reliably at close to real-time speeds. However, as resolution increases this becomes infeasible due to excessive storage and computational requirements. In this thesis it is proposed to reduce the textural components by modeling the coefficients of a wedgelet based regression tree instead of the original pixel intensities. The wedgelet regression trees employed are based on the triangular domains and estimated using cross validation. The wedgelet regression trees serves to 1) reduce noise and 2) produce a compact textural description. The wedgelet enhanced appearance model is applied to a case study of human faces. Compression rates of the texture information of 1:40 is obtained without sacrificing segmentation accuracy noticeably, even at compression rates of 1:115 fair segmentation is achieved. For regularization of geometric property of the wedgelet decomposition a Markov Random Field method is introduced which improves the performance of the segmentation on the wedgelet enhanced appearance model.