Material Recognition for Efficient Acquisition of Geometry and Reflectance

In proceedings of Computer Vision - ECCV 2014 Workshops, Zurich, Switzerland, pages 321-333, Springer International Publishing, Sept. 2014
 

Abstract

Typically, 3D geometry acquisition and reflectance acquisition techniques strongly rely on some basic assumptions about the surface reflectance behavior of the sample to be measured. Methods are tailored e.g. to Lambertian reflectance, mirroring reflectance, smooth and homogeneous surfaces or surfaces exhibiting mesoscopic effects. In this paper, we analyze whether multi-view material recognition can be performed robust enough to guide a subsequent acquisition process by reliably recognizing a certain material in a database with its respective annotation regarding the reconstruction methods to be chosen. This allows selecting the appropriate geometry/reflectance reconstruction approaches and, hence, increasing the efficiency of the acquisition process. In particular, we demonstrate that considering only a few view-light configurations is sufficient for obtaining high recognition scores.

Keywords: Material Recognition, reflectance, set-based classification

Bibtex

@INPROCEEDINGS{weinmann-2014-SetbasedMaterialRecognition,
     author = {Weinmann, Michael and Klein, Reinhard},
      pages = {321--333},
      title = {Material Recognition for Efficient Acquisition of Geometry and Reflectance},
  booktitle = {Computer Vision - ECCV 2014 Workshops},
     series = {Lecture Notes in Computer Science},
     volume = {8927},
       year = {2014},
      month = sep,
  publisher = {Springer International Publishing},
   location = {Zurich, Switzerland},
   keywords = {Material Recognition, reflectance, set-based classification},
   abstract = {Typically, 3D geometry acquisition and reflectance acquisition techniques strongly rely on some
               basic assumptions about the surface reflectance behavior of the sample to be measured. Methods are
               tailored e.g. to Lambertian reflectance, mirroring reflectance, smooth and homogeneous surfaces or
               surfaces exhibiting mesoscopic effects. In this paper, we analyze whether multi-view material
               recognition can be performed robust enough to guide a subsequent acquisition process by reliably
               recognizing a certain material in a database with its respective annotation regarding the
               reconstruction methods to be chosen. This allows selecting the appropriate geometry/reflectance
               reconstruction approaches and, hence, increasing the efficiency of the acquisition process. In
               particular, we demonstrate that considering only a few view-light configurations is sufficient for
               obtaining high recognition scores.}
}