Filling the signed distance field by fitting local quadrics

We propose a method of filling unmeasured regions of shape models integrated from multiple measurements of surface shapes. We use the signed distance field (SDF) as shape representation that contains information of the surface normal along with the signed distance at the closest point on the surface from the sampling point. We solve this problem by iteratively fitting quadratic function to generate smoothly connected SDF. We analyzed the relationship between the quadratic coefficients and the surface curvature, and by using the coefficients, we evenly propagated the SDF so that it satisfies the constraints of the field. The proposed method was tested on synthetic data and real data that was generated by integrating multiple range images.