Roger HUBBOLD, University of Manchester, GB (coded UMAN in following)
Juan J.ANZA, LABEIN, BILBAO, ES (coded LABEIN)
Didier ARQUES, University of Franche-Comte, FR (coded UFC)
Daniele MARINI, University of Milan, IT (coded UMIL)
David BOYD, Rutherford Appleton Laboratory, GB (coded RAL)
Jean-Pierre MADIER, CRIL/INFOROP,Toulouse, FR (coded CRIL)
Jean-Claude GROSSETIE, JRC, ISPRA, EEC (coded JRC)
Ralf GRUBER, CSCS, Manno, CH (coded CSCS)

Objectives of the Network:

Network methodology has been to organize activity in three Working Groups (WG) with the following themes:

Working Group on Scientific Visualisation (WG-SV)

• objectives: distributed visualisation, 3D/volume rendering, graphical and object-oriented user interfaces

• members: CERFACS, UMAN, LABEIN, RAL, CSCS.

• objectives: parallel algorithms for image synthesis with ray tracing and radiosity

• members: UMAN, UFC, UMIL, JRC.

• objectives: algorithms and experience exchange in image processing, image compression, remote sensing

• members: CSCS, UMIL, RAL, CRIL, JRC.

Each Working Group has been coordinated by the first member in the list. Some cross-over activities resulted due to partial overlap of the themes.

Roles of the Participating Teams:

• CERFACS (also acting coordinator for contract with the Commission) has been participating mainly in Scientific Visualisation and user interface, and following other subgroup activities such as compression and wavelets, object-oriented paradigms and user interfaces.

• UMAN has been providing techniques and software to enable distributed image synthesis (distributed rendering protocol - DRP), and assistance with implementation of parallel software codes, especially on the KSR1 computer, with provision of parallel ray tracing (KRT)and volume rendering (KVR) codes for comparisons.

• LABEIN has been working upon scientific visualisation and user interface, in CAD/CAM, metal forming and fluid dynamics applications areas, exchanging experience and data with other partners.

• UFC has contributed software for 3D modelling (SMILE) and realistic rendering (ray tracing and radiosity codes), and designing criteria for evaluation and comparison of this software with that developed by the other teams.

• UMIL has been providing know-how and software tools for volume visualisation (X-EVA) and fractal image compression, running on parallel computers and integrating them in state of the art software (CRIL's VINCI).

• RAL has expertise in scientific visualisation, image processing and satellite imaging, and has been collaborating with other labs to tests compression and imaging methods upon its working remote-sensing data (ERS-1/ATSR) and validating implementations of algorithms and results.

• CRIL is an industrial company in teledetection, image processing and hardware platforms market. Working with other partners, its aims to propose, implement and validate image processing, tranqsforms and compressions algorithms in its commercial grade software, VINCI, to be made available to partners for a one-year period.

• JRC implemented various fast transform algorithms that are widely used and known in image synthesis and analysis.to be integrated in VINCI; also wavelet and fractal techniques have been studied and implemented.

• CSCS has expertise in and developed tools for scientific visualisation (MOLEKEL package) and remote visualization (MET++ extensions), image compression techniques (UNDINE), and remote-sensing techniques. Specifically the compression algorithm has been distributed to and used by several partners to accomplish their objectives (integration in VINCI, applied on ATSR satellite images).

Major results:

A principal objective was to use advanced tools for visualization of numerical results, particularly those obtained in CFD simulations of industrial problems, as well as for visualization and animation of results in simulation of metal forming processes. The activities in this field were related to exploration of new visualization techniques for CFD and for metal forming simulations, evaluation and application of different tools to industrial problems and generation of an R&D project for development and industrial application of visualization techniques.

Several visits by LABEIN to Rutherford Appleton Laboratory, University of Manchester and CSCS have contributed to a common understanding of the current state of the art in advanced visualization tools as well as to improve visualization and animation of current CAE systems for CFD and FEM simulation.

Enhancements and developments of graphical user interfaces for the scientific visualization of molecular data have been realized. Most functional parts of the molecular graphics package MOLEKEL have been enhanced with an X/Motif compliant graphical user interface, and the rendering tasks have been ported to OpenInventor. An input generation tool for Gaussian-92, an intensively used computational chemistry package, has been developed to allow scientists to

setup and launch a numerical calculation from the same program that is afterwards used to analyze and visualize the results.

CERFACS have worked on 3D visualization and compression of volume data, for several applications. Information has been exchanged with University of Manchester and RAL. CERFACS originally intended to offer CS2 resources for collaborative work, but delays in the arrival and availability of this system meant that this could not be achieved.

Comparison of industrial GUI-builders and evaluation of own-developed builder tool were made by LABEIN and CSCS, and similar work is ongoing for public-domain tools.Virtual Reality became of significant interest to several partners (RAL, LABEIN, UMAN), with visits to partners and others laboratories. Potential applications include aerospace design, large scale CAD, and numerical simulation results visualisation.

The SMILE modelling language developed at UFC was distributed to UMAN and JRC. This provides a common 3D modeller, allowing standard scenes to be used with the different partners' software for comparison of algorithms and alternative parallel machines. A SMILE interface to KRT was developed at UMAN for this purpose.

The UFC ray-tracer was implemented on the KSR1 parallel computer at UMAN, permitting comparison of results with the KRT system, using the same test scenes. The UMIL ray tracer was also parallelised on the KSR1 at UMIL, and a paper published about this. Work on benchmarking techniques for ray-tracing is continuing beyond the formal end of the collaboration; a joint technical report has been produced detailing the comparisons made so far.

UFC and JRC have jointly developed new radiosity techniques, leading to much faster solution times. A paper on this work has been accepted for publication. The X-EVA volume rendering system (UMIL) was distributed to the other partners. It has been ported to the KSR1 at UMAN and a parallel version was implemented at CSCS. UMAN and UMIL worked together on refining the distributed rendering protocol (DRP) and integrating this into the X-EVA interface, providing a powerful platform-independent rendering environment. Interactive use of distributed rendering with this system (ray tracing and volume rendering) was demonstrated at the final project meeting at CSCS. Tests of DRP performance over the Internet were made from CSCS to Manchester; results of this are given in a paper submitted for publication.

Different ray tracers were ported on KSR1 system, providing the conditions to compare ray tracing on different architectures and from different algorithmic approaches, in particular those by UMAN and UFC teams. The effective comparison has not been completed within the network timing period.

JRC has worked on Fourier Hadamard, Paley, Walsh and Haar Transforms. All these transforms can be implemented by a Cooley-Tukey algorithm, or a modified Cooley-Tukey algorithm. Thus, computation of Fourier, Paley, Walsh, Hadamard and Haar transforms is fast and simple. JRC has also worked in the field of high compression fileds and wavelet and fractal analysis; compression-decompression is done with following steps:

• Define the eigen wavelets associated with a specific image.
• Compress the associated "image" using a block iterative scheme using eigen - wavelets.
• Define the eigen operator for associated quantification.
• Apply the advance lossless compression techniques for having the final result (LZW, Huffman, arithmetic coding ...). During the previous step, the fractal dimension and associated directory are computed, Thus adaptive wavelet techniques can be defined for the reconstruction.
• The reconstruction process is performed in the reverse order except the quantification.
• During the reconstruction process, fractal and eigen wavelets dictionary are updated.
• Finally the original image has been reconstructed but enlargement and zooming are now available in the same way as the fractals.

JRC has also worked in holographic synthesis, which requires an important amount of computational time as well as a large amount of storing capacity of computer peripheral devices. This is a consequence of extensive use of numerical fast transforms (like: FFT, Walsh...) and advanced coding (like: Lohmann, Lee, Tarasov...).It has been demonstrated that for CGH, all of these algorithms are well suited to massive parallel computers since at least 1 GFlop is needed.

JRC has large experience and expertise in the areas of parallel computing applications (SUPERNODE and CONCERTO CS1 programming, ray tracing, radiosity, CGH) and in the domain of holographic synthesis (optical transforms, fast transforms including Fourier, Hadamard, Walsh-Paley, Haar, Wavelet transforms). Recent domains of applications include multi-speckle holograms of coherently illuminated objects using an iterative fast-transform phase retrieval algorithm to study spatial correlation of laser-speckle intensity measurements.

CSCS has developed better image compression tools than currently available (such as JPEG) and applied them to image data supplied by the other partners. A new wavelet-based image compression algorithm has been developed and implemented (UNDINE). Comparisons on various image data show that in general much higher compression ratios that with JPEG can be achieved. The UNDINE code has been integrated in the VINCI image processing software of CRIL. The algorithm has been tested at RAL on remote sensing images (ERS-1/ATSR) and better understanding of the requirements of earth observation applications have been gained. A comprehensive technical report on wavelets has been published together with JRC Ispra.

Originally it was planned to embed further enhanced versions in a remote visualization environment. First steps lead to integration of remote video tools into a multimedia application framework. This will enable scientists without the necessary video facilities to generate and preview video animations from their remote site. This part has been realized at CSCS together with Multimedia Lab, University of Zurich.

A further activity at CSCS has been applications of multiresolution techniques in remote sensing. It envisages to give an overview of the field of multiresolution analysis and synthesis and mainly to establish bridges and to integrate different methods and techniques. The project itself has two parts: The first, Multiresolution Image Synthesis, presents a method for the accurate visualization of the corrugated surfaces, and a new technique for the topography influence rejection from optical Remote Sensing images, with applications in snow cover segmentation in alpine regions. The second part, Analysis of Multiresolution Stochastic Processes, contains and a report presenting the operatorial formalism of time, frequency, scale representations, its relations with the multiscale and multiresolution signal representation, and topics in hierarchic stochastic processes and their implications in multidimensional multiresolution signal characterization.

CSCS collected algorithms from various partners of the Network to produce an Algorithms Library that includes parallel image processing and synthesis modules developed for a distributed heterogeneous computer environment. The library is available for all partners of the project. It presently includes a color correction algorithm for laser printers, the parallelized volume visualization code X-EVA (UMIL), and the parallelized fractal image compression algorithm (UMIL). The implementation is based on PVM. Furthermore, the fractal image compression has been ported to the Cenju-3 massively parallel computer at CSCS and also been integrated into VINCI (CRIL).

CRIL Ingenierie integrated "ifs_comp" image compression algorithm into VINCI image processing system, allowing to use a powerful environment to explore different image compression schemes.

The commercial-grade VINCI software has been further enhanced and the following actions have been taken:

• Full technical support for the integration of transforms and compression algorithms
• Completion of the translation of the VINCI graphical user interface
• Edition of a VINCI documentation in PostScript
• Final tests of the enhanced VINCI on various types of images.

As some partners only own workstations of type Silicon Graphics (SGI), CRIL has migrated as part of this project the enhanced VINCI to SGI. This version has been actually presented at the final network meeting.

Comparison between objectives and work done:

• volume data manipulation:

  all the teams share common software tools and systems to explore volume data; moreover advanced architectures have been explored and special software components are now available to share freely computational resources both across WAN and local workstation networks;

• photorealistic image synthesis:

  a common knowledge is now available to all the teams, particularly concerning ray tracing and radiosity approaches; both approaches have been explored and compared on a wide set of computer platforms, including transputer based and shared memory architectures; the participating teams are now at the leading edge in this research field; knowledge and experience is also available in the domain of graphical user interfaces and could have been applied immediately to a molecular graphics application.

• image coding and compression:

  the possibility of studying and comparing different coding methods (JPEG, wavelets, IFS) have given the team a common knowledge on this field, and this knowledge is also integrated in a large software system.

Assessment of benefit working together

• Assessment of network contribution to training and mobility:

  CRIL employed on a temporary contract Guillermo Ciscar, spanish national, for working on VINCI in the framework of the CRIL-CSCS-JRC collaboration. G. Ciscar is currently working at JRC on image compression techniques for the VINCI environment.

  The University of Milano team has given a specific contribution based on the skills in the field of interactive volume visualisation, advanced photorealistic rendering and fractal based image compression. A series of seminars have been organised during the meeting on the following topics:

  • Volume manipulation and rendering methods

  • Computational methods for form factor estimation in radiosity based rendering algorithms

  • Parallelisation of ray tracing algorithms

  • Fractal image compression and comparison with JPEG and other methods.

  CSCS has given a tutorial, course and lecture on wavelets and their application to specific fields:

  • Tutorial on wavelets. Physics Computing `94 conference at Lugano (CH), August 1994.

  • Lecture on wavelet transforms and their application to image compression at Rutherford Appleton Labs, Didcot, September 1994.

  • Course on wavelet analysis and image compression at CRIL Toulouse, December 1994.

• Significant benefits from working together with other laboratories are:
  • Experience of international collaboration with partners of different nationalities and of different scientific background

  • A better knowledge of the state-of-the-art of image processing techniques, specifically in image compression and fast transforms

  • Development of strong relationships with some partners on which future collaborations could be built. As a first effect, two people previously at CRIL are now pursuing work at JRC.