Workshop on Geometry compression

Sophia-Antipolis
november 6, 10:00 to november 7, 17:00 2003

Venue information Program
Topics
  • compression of 3D objects
  • surface parameterization
  • video coding
  • watermarking of 3D object
Invited speakers
On the optimality of some 3D mesh compression methods
Markov Models for Optimal Arithmetic Coding

Arithmetic coding achieves provably optimal compression rates for all commonly used probability models, such as information sources with and without memory and the popular Markov Models. In a lot of applications, where an upper bound on the coding performance is necessary, the most difficult aspect is the design of a good probability model. This talk illustrates a new method to derive from a given state machine a Markov Model that encodes an output sequence from the state machine optimally in the limit for infinitely long sequences. With the new method the design problem is simplified to the design of a good state machine that enumerates all possible input sequences. For a commonly used triangle mesh connectivity coding method the design of such a state machine is shown and the application of the method allows for improving the upper bound for the coding performance.
Out-Of-Core Processing of Huge Polygonal Meshes

Organization
The workshop is organized by INRIA collaborative action Telegeo.

Registration

Talks

François Cayre (ENST-TSI)
Spatial watermarking for authentication of 3D triangles meshes
We present a new watermarking scheme dedicated to the authentication of 3D models. Our scheme is naturally invariant to translation, rotation and scaling of the model. Every other manipulation should erase the watermark. We show how to achieve near-optimal embedding of the hidden information using an indexed localization in the spatial domain. We also give several bounds of the algorithm, including the Minimum Watermark Segment (MWS), the actual length of the keys and the number of sites used to embed the watermark. Our results will show the relevance of the method against various manipulations.

Sébastien Valette, Alexandre Gouaillard and Rémy Prost (CREATIS)
Lossy and Lossless Multiresolution Mesh Compression via Wavelet Representation
We present a new scheme for decomposition of 3D meshes Geometry by Wavelet Decomposition. Based on an irregular subdivision scheme, this scheme applies directly to irregular meshes, without the need for remeshing. Associated with a compact connectivity scheme, we show how to achieve an efficient progressive compression of 3D meshes. Two approaches of compression are the described : progressive compression by geometry and connectivity refinement, and progressive compression by geometry refinement. We also present results obtained with both approaches on several reference models.

Stéphane Pateux, Nathalie Cammas (IRISA)
Meshes for video analysis and coding
We present a general approach for video analysis and representation with 2D deformable meshes. A hierarchical dynamic triangular mesh structure is estimated from the input image sequence through a global optimization scheme. Using wavelet decomposition, the triangular mesh embeds both texture and motion information along the video sequence. Such a representation can be used for efficient scalable video coding or for 3D estimation and modelisation.

Raphaèle Balter, Franck Galpin, Patrick Gioia, Luce Morin (IRISA)
3D Mesh processing for video coding
We present a video coding scheme based on 3D modelisation. From a video sequence of a static scene viewed by a moving camera, we extract a stream of textured 3D mesh models and camera positions. They are coded using a scalable representation and used at the decoder to regenerate the original video. Such a scheme allows very low bit-rate compression as well as virtual reality applications.

Benjamin Le Guen, Raphaèle Balter, Luce Morin (IRISA), Pierre Alliez (INRIA)
Morphing of estimated 3D models
We propose a morphing procedure between 3D mesh models, adapted to the case of estimated 3D models from a video sequence. Our goal is to obtain a visual continuity when reprojecting the 3D models onto the video camera viewpoints. We use 2D parametrisation and a mesh fusion procedure. We take into account the matching constraints provided by the motion field estimated from the image sequence.

Gilles Schaeffer (LIX)
Compact coding of triangulations, an enumerative approach.

Pierre Alliez (INRIA)
Anisotropic Polygonal Remeshing
We propose a novel polygonal remeshing technique that exploits a key aspect of surfaces: the intrinsic anisotropy of natural or man-made geometry. In particular, we use curvature directions to drive the remeshing process, mimicking the lines that artists themselves would use when creating 3D models from scratch. After extracting and smoothing the curvature tensor field of an input genus-0 surface patch, lines of minimum and maximum curvatures are used to determine appropriate edges for the remeshed version in anisotropic regions, while spherical regions are simply point-sampled since there is no natural direction of symmetry locally. As a result our technique generates polygon meshes mainly composed of quads in anisotropic regions, and of triangles in spherical regions. Our approach provides the flexibility to produce meshes ranging from isotropic to anisotropic, from coarse to dense, and from uniform to curvature adapted.

Tony Tung (ENST)
3D object indexing using Multiresolutional Reeb Graph
We present a 3D indexing method for retrieving 3D models of objects in dedicated databases. Among the various techniques in the literature, we have chosen the Multiresolutional Reeb Graphs (MRG) approach presented in [Hilaga et al, SIGGRAPH'01]. A Reeb graph is a unidimensional skeleton based on the topology of the object. The function used to build the MRG is robust in case of mesh modifications (noise on points, decimation, etc.). The comparison scheme is multiresolutional and uses a coarse-to-fine approach. We provided geometrical informations to the MRG to improve the matching of the graph for similarity estimation of 3D objects.

Guillaume Lavoué, Florent Dupont et Attila Baskurt (LIRIS)
Curvature Based Triangle Mesh Segmentation with Boundary Rectification
We present a new and efficient algorithm for surface decomposition of arbitrary triangle meshes, based on curvature information analysis. Our method not only "cut" the object along its edges but decomposes it into constant curvature surfaces, with a precision threshold given by the user. Contrary to the existing methods, a post-processing allow to obtain patches that have smooth and regular boundaries, and thus which can be later fit with NURBS or subdivision surfaces in a multi-resolution compression perspective.

Luca Castelli (LIX)
Canonical triangulation of a graph, with coding application

Frédéric Payan, Marc Antonini (I3S)
A Wavelet Geometry Coder including Normal Field Control
We present an original geometric wavelet coder for 3D meshes including a model-based bit allocation. The proposed coder takes into account the high sensitivity of the surface-to-surface distance to the normal information quantization. Indeed, the proposed bit allocation criterion is penalized by the knowledge of the polar angle between the quantization error vector and the normal direction in a local coordinate frame, and consequently permits the control of the error vector orientation. Then, bits are allocated preferentially to the normal set favorizing the minimization of a surface-to-surface distortion measure.

Carlos Hernandez Esteban (ENST)
Silhouette and Stereo Fusion for 3D Object Modeling
We present a new approach to high quality 3D object reconstruction. Starting from a calibrated sequence of color images, the algorithm is able to reconstruct both the 3D geometry and the texture. The core of the method is based on a deformable model, which defines the framework where texture and silhouette information can be fused. This is achieved by defining two external forces based on the images: a texture driven force and a silhouette driven force. The texture force is computed in two steps: a multi-stereo correlation voting approach and a gradient vector flow diffusion. Due to the high resolution of the voting approach, a multi-grid version of the gradient vector flow has been developed. A new formulation of the silhouette constraint is derived allowing its direct integration in the evolution algorithm. Finally we use the Laplacian and the biharmonic operators as internal forces which, together with a multi-resolution deformable model, allow us to obtain high quality 3D models.

Arnaud Gelas (CREATIS)
Curvature Based Remeshing of 3D Meshes
Multiresolution meshes are very attractive in many fields of research, such as Computer Graphics, Finite Element, and Medical Image Processing. There are two different ways to construct the mesh hierarchy, coarse to fine and fine to coarse methods. Our method is coarse to fine one, so it converts a mesh into a multiresolution one. Usual methods that are based on geometrical error, while the proposed method is an adaptive remeshing based on curvature information. We also show an interesting application of our remeshing method: compression of 3D Meshes.

Alexandre Gouaillard, Sebastien Valette, Remy Prost (CREATIS), in collaboration with Pierre Alliez (INRIA).)
Multiresolution mesh for 3D Segmentation
Segmentation of image and volumes proved to be more robust when an a-priori model is used. Previous work proved that using a multire'solution decomposition of the data AND of the model leads to 50 ( 3 levels of decompositions) to 66 % ( 5 levels of decomposition) theoretical decrease in time to converge to the same solution with the same Snake algorithm. Previous work showed that, on real case, the decrease was closer to 90 % and that the robustness was greatly improved in presence of noise.
Even if the same theoretical results can be achieved on 3D, building an adequate multire'solution mesh for segmentation is a difficult task. The number of vertices must stay as low as possible as the Snake algorithm's complexity depends linearly on it. On the other hand, the mesh must be as regular and uniform as possible for the discretization of the energy function that drives the model's deformation to stand. Last but not least, the multire'solution decomposition of the data tends to change the topology of the approximation of the object , while the multire'solution decomposition of a mesh never do. multire'solution decomposition of a mesh should allow some topological changes for the model and the data at a given resolution to be synchronised.
We will propose here a method to construct an as regular and uniform as possible multiresolution mesh for segmentation purposes. Discussion about how to include topological changes in a multiresolution decomposition algorithm will be provided as a "work under progress" part.

Patrick Gioia (France Telecom)
Real-time adaptive reconstruction of Wavelet Encoded Meshes
We present a set of algorithms for real-time decoding and reconstruction of meshes encoded using geometric wavelets. Wavelet compression of surfaces has proven to be a powerful tool for lossy mesh coding, and is now part of the MPEG-4 standard. While leading to impressive compression rates, the encoding relies on quite complex transformations which make real-time decoding and reconstruction not straightforward. The algorithms we propose allow view-dependent streaming and real-time adaptive reconstruction of meshes of arbitrary size. They can be used for scalable streaming of any medium with underlying meshed objects, such as model-based video or finite elements for motion compensation.

B. Levy (ISA), B. Vallet, A. Sheffer
Differentially Constrained Parameterization using Alternative Variables
These last few years, time and effort has been devoted to design parameterization algorithms for mesh models, minimizing area or angle deformations. Area-constrained and authalic parameterizations are best suited to map a signal onto a surface, like in texture mapping. Angle-preserving (i.e. conformal) parameterizations show suitable properties for remeshing algorithms and PDE solvers. We present two new parameterization methods that overcome some theoretical and practical limitations of previous work. Our authalic parameterization DPBF directly optimizes the Jacobian of the transform in dot-product space, which provides a finer control on the use of texture space. Our conformal parameterization ABF++ is a highly efficient extension of the ABF method. In ABF++, algebraic transforms dramatically reduce the dimension of the linear systems solved by ABF. Instead of directly optimizing $u,v$ coordinates, both DPBF and ABF++ use alternative variables, naturally adapted to the problem. To reconstruct the $u,v$'s from those variables, we propose LSAM, a new anisotropic generalization of LSCM. Using LSAM, the $u,v$'s are globally optimized, which avoids the numerical instabilities caused by accumulated errors in greedy algorithms, such as ABF's reconstruction phase.

Nicolas Ray (ISA)
Hierarchical Least Squares Conformal Maps
A texture atlas is an efficient way to represent information (like colors, normals, displacement maps ...) on triangulated surfaces. Our LSCM method (Least Squares Conformal Maps) automatically generates a texture atlas from a meshed model. For large charts (over 100k facets), the convergence of the numerical solver may be slow. It is well known that the conformality criterion, minimized by LSCM, also corresponds to a harmonicity condition, meaning that barycentric coordinates are locally preserved through the parameterisation. This has two different consequences:
o cascadic multigrid methods (coarse to fine) are well adapted to this criterion, and dramatically speed up the convergence of the numerical solver ;
o the obtained parameterisation naturally minimises texture swimming when used to texture-map a progressive mesh.
we introduce HLSCM (Hierarchical LSCM), a cascadic multigrid version of LSCM. As an example of possible applications, the paper shows how normal maps and simplified models can be automatically generated from large scanned meshes. Using these normal maps, the visual appearance of the model can be preserved more even when 90% of the vertices are removed from the initial model.

Marco Attene, Bianca Falcidieno, and Michela Spagnuolo (IMATI-GE / CNR - Italy), Jarek Rossignac (GVU Center / GATECH - Georgia-USA)
Remeshing-based compression of manifold triangle meshes
We focus on the lossy compression of manifold triangle meshes. Our SwingWrapper approach partitions the surface of an original mesh M into simply connected regions, called triangloids. From these, we generate a new mesh M'. Each triangle of M' is an approximation of a triangloid of M. By construction, the connectivity of M' is fairly regular and can be compressed to less than a bit per triangle using EdgeBreaker or one of the other recently developed schemes. The locations of the vertices of M' are compactly encoded with our new prediction technique, which uses a single correction parameter per vertex. SwingWrapper strives to reach a user-defined output file size rather than to guarantee a given error bound. For a variety of popular models, a rate of 0.4 bits/triangle yields an L2 distortion of about 0.01% of the bounding box diagonal. Due to their constrainded locations, the vertices of the new mesh M' are rarely aligned with possible sharp edges of the original mesh M. Consequently, such sharp edges are likely to be replaced by chamfer artifacts. We propose an automatic technique for restoring most of the original sharp edges and corners on M' without the need to encode any additional information.
Related papers: - Marco Attene, Bianca Falcidieno, Michela Spagnuolo and Jarek Rossignac, "SwingWrapper: Retiling Triangle Meshes for Better EdgeBreaker Compression", ACM Transactions on Graphics, 22(4), pp. 1-15, 2003.
- Marco Attene, Bianca Falcidieno, Jarek Rossignac and Michela Spagnuolo, "Edge-Sharpener: Recovering sharp features in triangulations of non-adaptively re-meshed surfaces", In Proceedings of the Eurographics / ACM SIGGRAPH Symposium on Geometry Processing, pp. 62-71, 2003.

Seungyong Lee (Postech, Korea)
Random-Accessible Mesh Compression
We present a novel approach to mesh compression, which we call random accessible mesh compression. The most notable benefit of this approach is the random-accessibility of the encoded mesh, which enables effective handling of large meshes with limited memory space.

Alex Yvart (LMC-IMAG)
Hierarchical Modeling of Smooth Surfaces of arbitrary topology
Two opposite schemes can be used in order to compute a smooth surface over a mesh : parametric surfaces or subdivision surfaces. Each method has its own advantages and drawbacks. Our purpose is to cumulate the advantages of both schemes. To achieve this goal, we compute a parametric surface which is refinable to add details and deal with multiresolution. We obtain a hierarchical structure of smooth surfaces capable of level of details edition.


Sponsors