Scalable Compression of Dynamic 3D Meshes

TNT members involved in this project:

Animated 3D content is becoming an integral part of many applications. It is already employed in numerous domains ranging from video games, animation movies, CGI films, and special effects to scientific visualization and CAD. In all these domains exists an increasing demand for efficient storage of animated 3D content. Furthermore, transmission of animated 3D content over different types of access networks (like the Internet, local area networks, or mobile networks) using different types of receiving devices (like PCs, laptops, PDAs, and smart phones) gains increasing importance. This imposes additional requirements to compressed data, since it has to be adaptable to network transfer rates and end-user devices.

DynamicMesh
Figure 1: A dynamic 3D mesh: a sequence of static meshes with same connectivity but time-varying geometry.

The goal of this project is to develop an efficient compression scheme for dynamic 3D meshes. The compression scheme should fulfill the following requirements. It should

  • be linear in runtime
  • create a compact bit stream (small storage space)
  • create a bit stream which allows successive decoding of frames or groups of frames
  • support spatial as well as temporal scalability.

Connectivity is encoded once in the beginning of the encoding process. Based on this connectivity at the encoder and decoder a layered decomposition is derived using mesh simplifications. This produces a hierarchical mesh which is exploited to encode vertex positions of each frame using spatial layers. This allows frame-wise reconstruction from coarse to fine at the decoder. Dependencies between spatial layers of the same frame and neighboring frames are exploited for efficient compression using a closed loop DPCM. Encoded spatial layers of the current frame and encoded spatial layers of reference frames are exploited for prediction of vertex positions, which are part of the currently encoded spatial layer. Obtained prediction errors of each frame are also organized in spatial layers. For each spatial layer they are encoded using separate entropy encoders. This allows spatial scalability. Furthermore, frames have not to be encoded in display order. Other frame orders like hierarchical B-frame order are allowed, which provides additionally to spatial scalability also temporal scalability.

Coder architechture
Figure 2: The coding architecture.

This coding architecture is adopted in the MPEG-4 standard which is referred to as MPEG-4 FAMC (MPEG-4, Part 16, Amd. 2, Frame-Based Animated Mesh Compression).

Show all publications
  • Nikolce Stefanoski, Jörn Ostermann
    Spatially and Temporally Scalable Compression of Animated 3D Meshes with MPEG-4/FAMC
    ICIP '08 - IEEE International Conference on Image Processing, San Diego, USA, October 2008
  • K. Mamou, N. Stefanoski, T. Zaharia, J. Ostermann, F. Prêteux
    Frame-Based Compression of Animated Meshes in MPEG-4
    ICME '08 - IEEE International Conference on Multimedia & Expo 2008, Hannover, Germany, July 2008
  • Nikolce Stefanoski, Libor Vasa, Jörn Ostermann
    Tutorial on Compression of 3D Meshes - Applications, Approaches, Standards
    3DTV-CON 08, Istanbul, May 2008