G. Bebis et al. (Eds.): ISVC 2009, Part I, LNCS 5875, pp. 644–655, 2009.
© Springer-Verlag Berlin Heidelberg 2009
Reducing Artifacts between Adjacent Bricks in
Multi-resolution Volume Rendering
Rhadamés Carmona
1
, Gabriel Rodríguez
1
, and Bernd Fröhlich
2
1
Universidad Central de Venezuela, Centro de Computación Gráfica,
1041-A, Caracas- Venezuela
rhadames.carmona@ciens.ucv.ve, gaborodriguez@gmail.com
2
Bauhaus-Universität Weimar, Fakultät Medien, 99423 Weimar, Germany
bernd.froehlich@uni-weimar.de
Abstract. Multi-resolution techniques are commonly used to render volumetric
datasets exceeding the memory size of the graphics board, or even the main
memory. For these techniques the appropriate level of detail for each volume
area is chosen according to various criteria including the graphics memory size.
While the multi-resolution scheme deals with the memory limitation, distracting
rendering artifacts become noticeable between adjacent bricks of different lev-
els of detail. A number of approaches have been presented to reduce these arti-
facts at brick boundaries, including replicating or interpolating data between
adjacent bricks, and inter-block interpolation. However, a visible difference in
rendering quality around the boundary remained, which draws the attention of
the users to these regions. Our ray casting approach completely removes these
artifacts by GPU-based blending of contiguous levels of detail, which considers
all the neighbors of a brick and their level of detail.
1 Introduction
During the past years multi-resolution hardware-accelerated volume rendering has
been an important research topic in the scientific visualization domain. Due to the
progressive improvements of imaging devices such as tomographs and magnetic
resonators, the size of volume datasets continuously increases. Such datasets often
exceed the available memory of graphics processing units (GPU), and thus multi-
resolution techniques need to be employed to guarantee interactive frame rates for
GPU-based rendering approaches. Out-of-core techniques are required for even larger
datasets exceeding the main memory capabilities of regular desktop computers, which
are generated e.g. by scientific projects such as the visible human ® [1] and the time-
dependent turbulence simulation of Richtmyer-Meshkov [2].
While multi-resolution approaches in combination with out-of-core techniques deal
with the memory limitations, distracting rendering artifacts between adjacent blocks
of different level of detail occur (see Fig. 1). The source of these visual artifacts is the
interpolation process since coarser data generates different samples during interpola-
tion, which may be mapped to different colors by classification and during integration
along the ray. The presence of these kinds of artifacts in the resulting image subcon-
sciously draws the attention of the user to these regions of the volume instead of al-
lowing the user to focus on the actual data.