Multidimensional morphable models: A framework for representing and matching object classes

This thesis describes a flexible model for representing images of objects of a certain class, such as faces, and introduces a new algorithm for matching the model to novel images from the class. The model and matching algorithm are very general and can be used for many image analysis tasks. The flexible model, called a multidimensional morphable model, is learned from example images (called prototypes) of objects of a class. In the learning phase, pixelwise correspon-dences between a reference prototype and each of the other prototypes are first computed and then used to obtain shape and texture vectors associated with each prototype. The mor-phable model is then synthesized as a linear combination that spans the linear vector space determined by the prototypical shapes and textures. We next introduce an effective stochastic gradient descent algorithm that automatically matches a model to a novel image by finding the parameters that minimize the error between the image generated by the model and the novel image. Several experiments demonstrate the robustness and the broad range of applicability of the matching algorithm and the underlyin