Efficient Segmentation of 3D Face Reconstructions

The segmentation of the face in 3D reconstructions is a crucial processing step within 3D face recognition systems. At this early processing stage discarding other surface portions such as collars, hats or hairstyle shall reduce the amount of data. In contrast to other approaches, the proposed algorithm uses only the face geometry and is therefore robust with respect to lighting conditions or texture quality. Assuming the skin region of a face is locally flat and closed, a binary mask image is created. Morphology and a simple heuristic are applied on connected components to select and join appropriate components. The implementation is straight forward, yielding just a few parameters and copes the problem without training procedure. A proof of concept is given and results are shown for several cases, limitations of the approach are discussed.