Techniques for improving stereo depth maps of faces

This paper presents an improved technique for determining the 3D structure of the human face form stereo images. The approach targets specific regions of the face individually. An application of the approach to estimate the 3D structure of the nose is presented. Stereo matching is extended and applied to colour in addition to grayscale, which improves matching in a low texture image like parts of the skin of a human face. Left-right consistency check is used to determine erroneous disparity estimates. Prior knowledge of the nose structure is used in conjunction to improve the quality of the 3D nose estimates.