Three-dimensional Morphable Models (3DMM) are known to be valuable tools for both face reconstruction and face recognition. These models are particularly relevant in safety applications or Computer Graphics. In this thesis, contributions are made to address the major difficulties preceding and during the fitting process of the Morphable Model in the framework of a fully automated system.It is shown to which extent the reconstruction and recognition results depend on the initialization and what can be done to make the system more robust, e.g. against vague feature positions or occlusions of the face.
Based on the 3DMM, a fully automated algorithm is presented. Support Vector Machines (SVMs) and the Morphable Model of 3D faces are combined for reconstructing a textured 3D model of a face from a single photograph or a raw video stream. The SVM delivers a list of candidates for several facial feature positions, and these are evaluated using a novel criterion that is based on the Morphable Model and a combination of linear projections. To make the algorithm robust with respect to head orientation, this process is iterated while the estimate of pose is refined. Finally, the feature points initialize the model-fitting procedure of the Morphable Model to result in a high-resolution 3D surface model.
Furthermore a new approach called self-adapting feature layers (SAFL) is presented. The algorithm integrates feature detection into the iterative analysis-by-synthesis framework, combining the robustness of feature search with the flexibility of model fitting. Templates for facial features are created and updated while the fitting algorithm converges, so the templates adapt to the pose, illumination, shape and texture of the individual face. The benefit of the proposed method is an increased robustness of model fitting with respect to unavoidable errors in the initial feature point positions. Such residual errors usually create problems when feature detection and model fitting are combined to form a fully automated face reconstruction or recognition system. Several case studies show the benefits of the new concept proposed in this work.
In addition to the SAFL concept, another focus of this work is the importance of contour information for the entire reconstruction result. The overall shape of the face is not exclusively determined by the contour line, but it is substantially influenced by it. Different approaches are presented improving the contour fitting of the 3DMM.
Besides initialization, robustness and better contour adaption, this thesis also addresses the problem of occlusions. Manually or automatically identified occlusions can be marked to exclude them from the fitting process. Different kinds of automated detection algorithms using the 3DMM, for different kinds of occlusions, are presented.
Finally, capabilities have been investigated for the reconstruction based on multiple images. Here the focus was not only on a better reconstruction of the overall shape, but on distinguishing features, such as wrinkles, birthmarks or freckles. If only one input image is used, these get poorly reproduced. The reconstruction out of multiple images enhances their reconstruction.
The essential approaches are applicable to other model- based approaches to image analysis and they include a number of general strategies to analysis-by-synthesis besides their contribution to the improvement of the 3DMM.