Vector Solid Textures
Lvdi Wang
1,4
1
Tsinghua University
Kun Zhou
2
2
Zhejiang University
Yizhou Yu
3
3
University of Illinois at Urbana-Champaign
Baining Guo
1,4
4
Microsoft Research Asia
Figure 1: Several vector solid textures rendered in real time. In each group, the upper inset shows a closer view while the lower inset is
rendered using the bitmap solid texture from which the vector version is generated. In the middle group, the reflectance coefficients are stored
along with the RGB channels. In the rightmost group, three different scales of the same vector solid texture are composited to yield complex
self-similar boundaries.
Abstract
In this paper, we introduce a compact random-access vector repre-
sentation for solid textures made of intermixed regions with rel-
atively smooth internal color variations. It is feature-preserving
and resolution-independent. In this representation, a texture vol-
ume is divided into multiple regions. Region boundaries are im-
plicitly defined using a signed distance function. Color variations
within the regions are represented using compactly supported ra-
dial basis functions (RBFs). With a spatial indexing structure, such
RBFs enable efficient color evaluation during real-time solid texture
mapping. Effective techniques have been developed for generating
such a vector representation from bitmap solid textures. Data struc-
tures and techniques have also been developed to compactly store
region labels and distance values for efficient random access during
boundary and color evaluation.
Keywords: Vector Images, Solid Textures, Texture Synthesis
1 Introduction
Texture maps used for real-time rendering are traditionally stored as
bitmap images on graphics cards for random access. There has been
a constant tension between texture resolution and memory usage.
An overly fine resolution consumes too much GPU memory while
an insufficient resolution leads to blurry interpolated texture map-
ping results. This is even worse for solid textures because of large
memory consumption by 3D grids. A mipmap facilitates texture
minification, but not magnification. On the other hand, vector im-
ages are well-suited for magnification because of their resolution-
independent representation. The same vector image can be raster-
ized to a high-quality bitmap image at a wide range of resolutions.
It seems promising to solve the above dilemma for solid textures by
developing a specifically tailored vector representation.
Achieving this goal imposes a few challenging requirements. First,
the vector representation should be feature-preserving during mag-
nification. Features refer to edges where there exist sharp color or
intensity changes. In a solid texture, such edges typically form sur-
faces within the texture volume. Second, the vector representation
needs to be reasonably compact to alleviate memory consumption.
A vector representation becomes advantageous only when it con-
sumes a relatively small amount of memory but delivers a quality
equivalent to a relatively high resolution bitmap. Third, the vec-
tor representation should support fast random access and real-time
solid texture mapping on surfaces. Otherwise, it cannot become a
replacement for bitmap textures.
In this paper, we introduce an effective vector representation for
solid textures to meet the aforementioned requirements. It has the
following important characteristics. First, it decomposes a texture
volume into nonoverlapping regions along texture features. Re-
gion boundaries are implicitly defined using a signed distance func-
tion. Such a decomposition enables the preservation of rapid color
changes across features. Second, color variations within the regions
are compactly represented using radial basis functions (RBFs) with
a finite support. Such RBFs support efficient color evaluation dur-
ing real-time solid texture mapping. Third, data structures and
techniques have been developed to compactly store region labels
and distance values for efficient random access during boundary
and color evaluation. Furthermore, this representation facilitates
region-based texture composition and real-time texture editing op-
erations, including parametric warp and local boundary softness.
We have developed effective techniques for generating such a vec-
tor representation from either 2D exemplars or existing solid tex-
tures, and for mapping such vectorized solid textures onto mesh