Methods for 3D Geometry Processing in the Cultural Heritage Domain

Dissertation, Universität Bonn, May 2007
 

Abstract

This thesis presents methods for 3D geometry processing under the aspects of cultural heritage applications. After a short overview over the relevant basics in 3D geometry processing, the present thesis investigates the digital acquisition of 3D models. A particular challenge in this context are on the one hand difficult surface or material properties of the model to be captured. On the other hand, the fully automatic reconstruction of models even with suitable surface properties that can be captured with Laser range scanners is not yet completely solved. This thesis presents two approaches to tackle these challenges. One exploits a thorough capture of the object’s appearance and a coarse reconstruction for a concise and realistic object representation even for objects with problematic surface properties like reflectivity and transparency. The other method concentrates on digitisation via Laser-range scanners and exploits 2D colour images that are typically recorded with the range images for a fully automatic registration technique.

After reconstruction, the captured models are often still incomplete, exhibit holes and/or regions of insufficient sampling. In addition to that, holes are often deliberately introduced into a registered model to remove some undesired or defective surface part. In order to produce a visually appealing model, for instance for visualisation purposes, for prototype or replica production, these holes have to be detected and filled. Although completion is a well-established research field in 2D image processing and many approaches do exist for image completion, surface completion in 3D is a fairly new field of research. This thesis presents a hierarchical completion approach that employs and extends successful exemplar-based 2D image processing approaches to 3D and fills in detail-equipped surface patches into missing surface regions. In order to identify and construct suitable surface patches, selfsimilarity and coherence properties of the surface context of the hole are exploited.

In addition to the reconstruction and repair, the present thesis also investigates methods for a modification of captured models via interactive modelling. In this context, modelling is regarded as a creative process, for instance for animation purposes. On the other hand, it is also demonstrated how this creative process can be used to introduce human expertise into the otherwise automatic completion process. This way, reconstructions are feasible even of objects where already the data source, the object itself, is incomplete due to corrosion, demolition, or decay.

(Thesis submission date: Dec. 2006)

Download: http://nbn-resolving.de/urn:nbn:de:hbz:5N-10865

Bibtex

@PHDTHESIS{bendels-2007-dissertation,
    author = {Bendels, Gerhard H.},
     title = {Methods for 3D Geometry Processing in the Cultural Heritage Domain},
      type = {Dissertation},
      year = {2007},
     month = may,
    school = {Universit{\"a}t Bonn},
  abstract = {This thesis presents methods for 3D geometry processing under the aspects of cultural heritage
              applications. After a short overview over the relevant basics in 3D geometry processing, the present
              thesis investigates the digital acquisition of 3D models. A particular challenge in this context are
              on the one hand difficult surface or material properties of the model to be captured. On the other
              hand, the fully automatic reconstruction of models even with suitable surface properties that can be
              captured with Laser range scanners is not yet completely solved. This thesis presents two approaches
              to tackle these challenges. One exploits a thorough capture of the object’s appearance and a
              coarse reconstruction for a concise and realistic object representation even for objects with
              problematic surface properties like reflectivity and transparency. The other method concentrates on
              digitisation via Laser-range scanners and exploits 2D colour images that are typically recorded with
              the range images for a fully automatic registration technique. 
              
              After reconstruction, the captured models are often still incomplete, exhibit holes and/or regions
              of insufficient sampling. In addition to that, holes are often deliberately introduced into a
              registered model to remove some undesired or defective surface part. In order to produce a visually
              appealing model, for instance for visualisation purposes, for prototype or replica production, these
              holes have to be detected and filled. Although completion is a well-established research field in 2D
              image processing and many approaches do exist for image completion, surface completion in 3D is a
              fairly new field of research. This thesis presents a hierarchical completion approach that employs
              and extends successful exemplar-based 2D image processing approaches to 3D and fills in
              detail-equipped surface patches into missing surface regions. In order to identify and construct
              suitable surface patches, selfsimilarity and coherence properties of the surface context of the hole
              are exploited. 
              
              In addition to the reconstruction and repair, the present thesis also investigates methods for a
              modification of captured models via interactive modelling. In this context, modelling is regarded as
              a creative process, for instance for animation purposes. On the other hand, it is also demonstrated
              how this creative process can be used to introduce human expertise into the otherwise automatic
              completion process. This way, reconstructions are feasible even of objects where already the data
              source, the object itself, is incomplete due to corrosion, demolition, or decay.},
       url = {http://nbn-resolving.de/urn:nbn:de:hbz:5N-10865}
}