(The image is by courtesy of Andrei Khodakovsky, Peter Schröder and Wim Sweldens, CalTech)
Geometry Compression
Sam Liang
In addition, with the rapid development of the Internet, more and more graphics applications are running across the Internet, such as distributed collaborated modeling and multi-user games. It is essential for such distributed applications to be able to transmit geometry efficiently, which in turn require good geometry compression schemes.
Furthermore, the speed gap between CPU and memory is becoming even larger. Interactive animation of geomtric models require that large amount of triangles be efficiently loaded into the on-chip cache across the bus. Current bottleneck is the memory bandwidth, which also craves for good geometry compression algorithms to reduce the traffic to the slow memory.
This web page collects some information on Geometry Compression, including
some important papers, well-known researchers, useful web links, and other
relevant information.
Geometry: Vertex positions.
Connectivity: Specifies the triangles.
Properties: Colors, normals, etc.
Compressed form of the three components in the input.
Multiresolution Progressive Compression: Hoppe, Taubin, et al., Khodakovsky, et al. Prediction of vertex positions is performed in a hierarchical fashion. Connectivity bits are around 4-10bits per vertex.
Lossless vs. Lossy Compression.
Arbitrary Meshes vs. Canonical Meshes.
Static Meshes vs. Dynamic Meshes.
2. Mike Chow, "Optimized
Geometry compression for real-time rendering", Proceedings on the IEEE
Visualization'97.
Mike Chow presents an algorithm to efficiently produce generalized
triangle meshes. Their meshifying algorithms and the variable compression
method achieve compression ratios of 30 and 37 to one over ASCII encoded
formats and 10 and 15 to one over binary encoded triangle strips. Their
experimental results show a dramatically lowered memory bandwidth required
for real-time visualization of complex datasets.
3. Gabriel Taubin and Jarek
Rossignac, "Geometric
Compression Through Topological Surgery", ACM Transactions on Graphics,
Vol. 17, No.2, April 1998, pp.84-115.
In their algorithms, vertex positions are quantized within the desired
accuracy, a vertex spanning tree is used to predict the position of each
vertex from its ancestors in the tree, and the correction vectors are entropy
encoded. Properties, such as normals, colors, and texture coordinates,
are compressed in a similar manner. The connectivity is encoded with no
loss of information to an average of less than two bits per triangle.
Look at Jarek Rossignac's publication
page for more papers by him.
4. Gabriel Taubin, Andre Gueziec, Willian Horn, Francis Lazarus. "Progressive
Forest Split Compression". Proc. Siggraph 1998, pp. 123-132.
This paper introduces a new adaptive refinement scheme for string and
transmitting manifold triangular meshes in progressive and highly compressed
form. It achives high compression rate -- a forest split operation doubling
the number n of triangles of a mesh requires a maximum of approximately
3.5n bits to prepresent the connectivity changes. The paper also shows
how any surfaces simplifcation algorithm based on edge collapses can be
modified to convert single resolution triangular meshes to the PFS format.
Click here to see a more detailed critique
by Sam Liang.
5. Touma and Gotsman, "Triangle
Mesh Compression", Graphics Interface'98. Vancuver, Canada.
Mesh connectivity is encoded in a lossless manner, based on properties
of planar graphs. Vertex coordinate data is quantized and then losslessly
encoded using elaborate prediction schemes and entropy coding. Vertex normal
data is also quantized and then losslessly encoded using prediction methods.
6. Stefan Gumhold. "Real
Time Compression of Triangle Mesh Connectivitity". Siggraph'98.
Their algorithm has some nice properties, such as: i) simple implementation,
potentially in hardware; ii) very fast compression speed: 750,000 triangles
per second on Pentium with 300 MHz; iii) Even faster decompression: 1,500,000
triangles per second; iv) Connectivity compression to about 1.6 bits/triangle
(Huffman encoding) or 1.0 bits/triangle (adaptive Arithmetic coding).
7. S. Gumhold and Klei, "Compression of discrete multiresolution models", Technical Report WSI-98-1, Wilhelm-Schickard-Institut für Informatik, University of Tübingen, Germany, January 1998.
8. J. Snoeyink and M. Kreveld. "Linear-Time
Reconstruction of Delaunay Triangulations with
Applications".
Proc. European Symposium on Algorithms, 1997.
Their method can be seen as a form of geometry compression and progressive
expansion. Using their method, the structure of a terrain model can be
encoded in a permutation of the points, and can be reconstructed incrementally
in linear time.
9. J. Lengyel.
"Compression
of Time-Dependent Geometry", ACM 1999 Symposium on Interactive 3D Graphics.
This Paper presents a new view of time- dependent geometry as a streaming
media type and presents new techniques for taking advantage of the large
amount of coherence in time.
10. Hugues Hoppe.
"Progressive
Meshes". Proc. Siggraph 1996, pp. 99-108.
This paper presents the progressive mesh (PM) representation, a new
scheme for stroing and transmitting arbitrary triangle meshes. This is
a lossless, continuous-resolution representation, and it addresses several
practical problems in graphics: smooth geomorphing of level-of-detail approximations,
progressive transmission, mesh compression, and selective refinement.
11. Andrei Khodakovsky, Peter Schröder, Wim Sweldens. "Progressive
Geometry Compression". Accepted by Siggraph 2000.
They propose a new progressive compression scheme for arbitrary topology,
highly detailed and densely sampled meshes arising from geometry scanning.
They observe that meshes consist of three distinct components: geometry,
parameter, and connectivity information. The latter two do not contribute
to the reduction of error in a compression setting. Using semi-regular
meshes, parameter and connectivity information can be virtually eliminated.
Coupled with semi-regular wavelet transforms, zerotree coding, and subdivision
based reconstruction they see improvements in error by a factor four (12dB)
compared to other progressive coding schemes.
