Gap-Sensitive Segmentation and Restoration of Digital Images

Many methods exist for removing defects such as gaps, cracks, and disconnections from digital shapes. However, most such methods have several limitations, such as removing both erroneous and important shape details, or requiring non-trivial effort from the end user in the form of manual delineation or parameter setting. In this paper, we propose a technique for removing defects such as internal gaps and cracks from 2D and 3D digital shapes. For this, we first classify gaps as boundary detail (to be preserved) and interior errors (to be removed), based on a heuristic that uses the gap position with respect the medial axis of the simplified shape. Next, we remove error gaps using an efficient distance-based filling. We illustrate our method on robust segmentation and hair removal tasks for skin imaging, and compare our results with a number of relevant techniques in this area.