Download - Academy Publisher

ISBN 978-952-5726-09-1 (Print)
Proceedings of the Second International Symposium on Networking and Network Security (ISNNS ’10)
Jinggangshan, P. R. China, 2-4, April. 2010, pp. 254-257
An Improved Volumetric Compression
Algorithm Based on Histogram Information
Liping Zhao, and Yu Xiang
School of Mathematics and Information Engineering, Jiaxing University, Jiaxing 314001, China
Email: zhaoliping_jian@126.com
Abstract—Considering the histogram information can
reflect the statistical characteristics of volume data ,an
improved volume data compression algorithm based on
classified VQ is presented. Firstly, statistic the
characteristics of the data itself using histogram technique,
and classify the blocks which divided from total volume data
into two groups, the blocks with meaningless information as
one group(also called empty blocks), and those with
meaningful information as the other group(also called
object blocks). Secondly object blocks are then compressed
by vector quantized. When applying this algorithm to the
volume data, all experimental results demonstrate the
proposed algorithm can improve the compression rate
significantly in the premise of the good quality of
reconstruction image.
Index Terms—Vector quantization, Classify, Object blocks,
Volume compression
I. INTRODUCTION
Volume rendering [1] is the process of projecting a
volumetric three-dimensional dataset onto a
two-dimensional image plane in order to gain some
meaningful information about the dataset. For a large
volumetric dataset, its real-time volume rendering using
hardware acceleration largely depends on and often gets
limited by the capacity of graphics memory. Accordingly,
CVR (Compressed Volume Rendering) [2-5], with the
principle of coupling decompression with rendering, has
been proposed and has shown to be an effective approach
for solving the just mentioned problem.
Vector quantization [2-5], hereafter abbreviated to VQ,
is an ideal choice as an asymmetric coding scheme for
CVR. Although the encoding of VQ is complex, its
decoding is simple because it is essentially a single table
look-up procedure. VQ has first been applied in volume
rendering for compression purpose by [2]. A hierarchical
VQ method has been described in [3]; it can lead to good
fidelity, but it has lower compression rate by about three
times compared with the classified VQ method in [2]. In
[4], a volumetric compression algorithm has been
developed based on transform coding by exploiting
Karhunen-Loeve-Transformations and classified VQ,
where each block is classified by the contribution of
regions in the transform domain to the fidelity of the
reconstructed block. However, such an algorithm is rather
complicated and time-consuming. We have presented an
efficient volumetric compression algorithm called
FCHVQ (Flag based Classified Hierarchical VQ) in [5].
FCHVQ adopts a subtle classification strategy and sets
flags for each block during the data pre-processing stage.
© 2010 ACADEMY PUBLISHER
AP-PROC-CS-10CN006
254
Note that for most volumetric datasets, they are not
chaotic, and usually they have a certain percentage of
empty regions. Moreover, when the block size is not too
large, data in one block are usually highly correlated. As
a result, applying FCHVQ to volumetric compression can
get good performance.
In the context of CVR, compression rate and decoding
speed are the key concerns. Taking into account the fact
that histogram information reflects statistics of a
volumetric dataset, an improved volumetric compression
algorithm is presented in this paper. With the objective of
obtaining higher compression rate while at the same time
resulting in satisfactory final image, our algorithm
classifies the blocks into two groups based on histogram
information of the volumetric dataset: empty blocks and
object blocks. Most experiments show that more than
60% of blocks are empty blocks. Object blocks are then
vector quantized to achieve higher compression rate.
II. RELATED WORKS
Regarding image compression, it should be pointed out
that some of the blocks contribute more to the visual
quality of the final image, while the others make little
contributions. To address this issue, all the blocks need to
be classified into different groups, with each group being
quantized separately, thereby guaranteeing their presence
in the code vector population. Classification of blocks is
more important in the case of volumetric compression
due to arbitrary nonlinear mapping of the transfer
function. In addition, classification of blocks allows the
encoder to adapt to a volumetric dataset containing
various perceptually important features (e.g., surfaces,
boundaries, etc.). Therefore, most literature works [2-5]
adopt the idea of classification for the goal of obtaining
better performance. In the following we will provide a
summary of the relevant background work.
A. Classified VQ
Classified VQ was introduced in [2]. The motivation
was to speed up volume rendering of scalar fields through
ray tracing. Different from the simplest material model, a
shading model sensitive to surface normal was used. A
tuple { f , nx , ny, n
z}
was encoded at each voxel, where
f stands for the scalar field value, and ( nx , ny, n
z)
is
determined by the direction of the field gradient. The
classification scheme simply defined two groups of
blocks: uniform blocks and active blocks. Uniform blocks,
with the size of each being 2×2×2, are those in which