Context and motivations 
The context of the thesis subject is in line with the subject proposed and carried out a few months aga by Xun Gong. The latter was oriented to the exploitation of the accumulation based algorithm [1, 2] with the goal of mesh segmentation in link to the identification of tubular parts and its consecutive processing with objective of mesh smoothing of reconstruction if we consider partial mesh or point clouds with partial density. This PhD subject relies on the obtained new results of segmentation based on the voxel accumulation and propose to exploit the resulting promising perspectives shownfrom the prototype code implemented by Xun Gong. 
As the master subject, this PhD proposai is motivated by the difficult problem of identification and separation of geometric structures that can be of key importance to simplify, edit or improve the quality of a mesh. The,sè research axes have concrete applications in video games where 3D shapes can be modeled from scans and then edited and re-meshed and eventually animated from a skeleton. ln parallel, recently the acquisition methods based on photogrammetry have been democratized and now allow to produce 3D reconstructions from a number of photos without ma nuai editing or complex parameters. 
Subject 
Based on the new segmentation algorithm defined in the voxel accumulation space [3], the starting point of the PhD will be extend the segmentation strategy by including more geometric constraint during the extension phase. Such constraints can bè related to the confidence definition or by exploiting the mean radius estimated locally in the accumulation space. The images (a) and (b) of figure 1 illustrates the key steps obtained from the segmentation algorithm. ln order to diffuse the results of the proposed approach, other experiments need to be exploited in particular to performs comparisons to other methods like the one based on volumic abject [4) or from curvature [5] or from geodesic distances [6]. 
ln the continuity of the previous research axes, new objective are proposed du ring the PhD: 
•    Adaptative denoising from geometric segmentation parts. This point was a first interesting perspective that emerge from the experiments supported by Xun who implemented a prototype by exploiting common tools based on the DGtal library and available on GitHub. This point will be a first objective with the idea of diffusing another online demonstration as done for the accumulation.
•    Adaptative shape remeshing from geometric features. Following the previous point, another result objective could to exploit the segmentation result in order to refine surface mesh to provide smoother shape representation.
•    Partial mesh reconstruction from accumulation. Another point resulting from the previous master internship development cou Id be to combine the previous strategy from accumulation voxel expansion in order to reconstruct shape from partial acquisition process. This part could rely for instance from symmetric constraints in order to orient the shape reconstruction.