Saliency detection for 3D surface reconstruction, segmentation and simplification
While the laser scanning systems usually have limited field of view, the captured data from a single viewpoint can only cover a part of the area of interest. In this case, the data captured from different viewpoints have to be registered and integrated for the removal of redundant information and construction of a single digital watertight model. This project attempts to develop techniques for the estimation of the extent to which different points are important for the definition of the geometry of the whole shape, so that only such points and patches with similar properties can be integrated, grouped together, or removed. To this end, some invariants need to be extracted and processed from the shapes under consideration.
Integration of various range scans. From left to right: volumetric method [Dorai and Wang 1998], mesh-based method [Sun et al. 2003], FCM [Zhou et al. 2009], k-means clustering [Zhou and Liu 2008], MRF-based method [Paulsen et al. 2010], Saliency-guided method.