From: Marchadier Jocelyn <jm@pr...>  20040503 17:13:05

Hello, I am Jocelyn Marchadier. I am a postdoc in the PRIP group at Vienna univers= ity=20 of technology. The two main topics studied at PRIP are graphbased image=20 analysis and 3D vision. Among the ongoing projects, we have basic research= =20 projects (irregular pyramids, cognitive vision) and application oriented=20 projects (geographic data representation, reconstruction of archeological=20 fragments or monitoring and surveillance activities). Further information c= an=20 be found at http://www.prip.tuwien.ac.at. I work on two distinct and complamentary fields: digital (and algebraic)=20 topology and functional modeling, applied to computer vision and pattern=20 recognition. I am particularly interested in producing structures from=20 digital images which can be transposed without ambiguity to the continuous= =20 domain and related problems like combinatorial optimisation for feature=20 extraction and matching. I recently proposed algorithms for computing=20 watersheds with nice topological properties and model based image encoding= =2E =20 My current envisionned applications are matching on stereo image sequences= =20 and generic object recognition (for cognitive vision). How I came to VXL ? Over the years, I have developed many tools for testing= my=20 CV algorithms (bundle adjustment, feature extraction, watersheds,=20 reconstruction, image loaders...). Many of them were just data structures,= =20 routines for computing geometric properties, etc. I was constantly needing= =20 basic tools (gaussian sample generation, random permutations, SVD, ...) and= I=20 was looking for them on the internet and integrating them until I discovere= d=20 the VXL homepage. I not only found many of the things I needed, but I also= =20 realized that at least 50% of my efforts had been useless, as they had=20 already been integrated in a coherent, complete and maintained library. As= I=20 had hundreds of unmaintained basic functions some of them having unresolved= =20 bugs ..So I decided to slowly migrate my most important sources to it. In PRIP, I studied and enhanced a representation which encode topological=20 pyramids (originally proposed by Luc Brun and Walter Kropatsch). Topologica= l=20 pyramids are irregular pyamids which can be seen as sequences of "partition= s"=20 derived from one another by merging adjacent faces or edges. The pyramid=20 structure that was developed at PRIP enable the representations of all the= =20 cells and their adjacency relations with a linear (!) memory complexity.=20 This representation and the related software have been used in different=20 projects, like geographical information segmentation/representation and=20 stereo sequences analysis. We think that this software could be a valuable= =20 contribution to VXL. So, we will propose as a contribution several structur= es=20 for handling topological representations and pyramids. I will later submit= =20 one of my recent watershed algorithm and the basic framework for segmenatio= n=20 using combinatorial maps and pyramids. Maintenance is also one of our=20 concern and we are looking forward to make more of our sources available. I am currently working on the vxlcompliance of the sources. I will submit= =20 them before the end of the week. This should appear in the contrib director= y=20 under "prip". I envisioned the names "vmap" and "vpyr" (which do not seem t= o=20 be taken) for two libraries, one manipulating combinatorial and topological= =20 maps, and the other extending the first one to pyramids. Please, tell me if these names are fine.=20 Best regards, =2D=20 Jocelyn Marchadier Pattern Recognition and Image Processing Group (PRIP) Institute for Computer Aided Automation Vienna University of Technology =46avoritenstra=DFe 9/1832 A1040 Wien, AUSTRIA tel. : (+431) 5880118358 email : jm@... 