Thesis of Florian Caillaud

Progressive and out-of-core compression of 3D generic meshes

Defense date: 17/01/2017

Advisor: Guillaume Lavoué
Coadvisor: Vincent Vidal


Technological advances in the fields of telecommunications , computer graphics and design of computer hardware over the past two decades have contributed to the emergence of digital three-dimensional (3D) data in many industries: Scientific visualization , digital entertainment, computer-aided design , cultural heritage or geographic information systems . These 3D data is typically represented by polygonal surface meshes. Indeed they are one of the main output representations in 3D creation algorithms, post - processing, analysis and reconstruction applied to scalar fields of varied origin , for example the results of scientific simulations. The increased complexity of these meshes, often enriched with attributes such as color , texture maps or various measures associated with the vertices , requires efficient compression methods to reduce their storage space and especially to reduce the time exchange during transmission across the network as part of a remote display , for instance.
Moreover, because of the still finer sampling measures , and increasing demands in terms of accuracy and / or quality rendering , 3D meshes have a strong tendency to become huge (hundreds of millions of elements, sometimes billions ) data compression should be able to do without all the data is loaded into RAM , is what commonly out- of-core method is called .

Scientific objectives
Among the various existing algorithms for compression, progressive compression techniques seem most relevant , especially in a scenario of remote viewing which requires not only a fast data transmission but also an adaptation complexity in terminal client. Many algorithms exist for the progressive compression. The issue of compression and , more generally, out- of-core processing 3D meshes has also been extensively studied. However, no method of out- of-core and progressive compression , treating generic polygonal meshes , has been proposed to date. We mean by generic 3D meshes with non manifold singularities, many related parts and texture maps and / or associated attributes , as the vast majority of 3D data used in the industry and the general public.
It will be crucial to take into account not only the geometry but also associated data in the proposed approaches . The visual quality of level of detail is a critical factor in our scenario of interactive remote visualization.
Another particularly important aspect in the case of very complex data integration is random access which is to be able to decompress only the parts of the 3D object or scene , for example to the viewing function of the camera position