In 3D surface scanning it is desirable to lter away bad data without altering the quality of the remaining good data. Filtering of raw scanner data before surface reconstruction can minimize the induced er- ror and improve on the probability of reconstructing the true surface. If outliers consist of actual data such as hair, and not just evenly dis- tributed noise, these outliers tend to err smoothing algorithms away from the wanted result. We present a novel algorithm based on a Markov Ran- dom Field that uses a distance constraint to robustly classify a 3D scan volume. Through this classication a signal preserving ltering of the data set is done. The remaining data are used for a smooth surface re- construction creating very plausible surfaces. The data used in our work comes from a newly developed hand held 3D scanner. The scanner is an Ultra Fast Optical Sectioning scanner, which is able to extract high quality 3D surface points from 2D images recorded at over 3000 fps. The scanner has been developed for digital impression taking in the dental area. Our work relates to future in-ear scanning for tting custom hearing aids without impression taking.
|Title of host publication||Proceedings of the MICCAI workshop on Mesh Processing in Medical Image Analysis (MeshMed)|
|Publication status||Published - 2011|
|Event||MICCAI workshop on Mesh Processing in Medical Image Analysis - Toronto, Canada|
Duration: 18 Sep 2011 → …
|Workshop||MICCAI workshop on Mesh Processing in Medical Image Analysis|
|Period||18/09/2011 → …|