Bill Triggs (Bill.Triggs@inrialpes.fr) and Matthijs Douze.
Keywords : digital photography, image correspondence, scene
reconstruction, image based rendering
Many people have large and growing digital photo collections. Finding the photo that you want can be difficult and there is a lot of interest in finding new ways to organise personal and shared photo collections. For popular tourist sites such such as Notre Dame or the Louvre hundreds if not thousands of images are available and it seems appropriate to try to organise them geometrically according to a rough floor plan of the scene, so that one can walk around in the virtual scene looking at the available images in much the same way as one walked around the real scene to take them. Achieving this automatically requires partial scene reconstruction based on robust visual correspondence to create an approximate scene model "annotated" with images, and an appropriate interface to walk through the model browsing the images.
The project will provide hands-on experience of using methods for visual correspondence, scene reconstruction and image based rendering. It will be based mainly on the methods described in the SIGGRAPH 2006 paper Photo Tourism : Exploring Photo Collections in 3D by N. Snavely, Steve Seitz and R. Szeliski. The LEAR team already has initial implementations of some of the building blocks required (notably the basic visual correspondence and partial scene reconstruction chains) but these will need to be adapted to the project and the image-enriched 3D model and navigation interface will need to be implemented.
Implementing the entire PhotoTourism system would be quite challenging and it will be necessary to select which parts to implement in conjunction with the candidate. The ideal candidate would have an engineering orientation and wish to go on to do a PhD in visual scene reconstruction or image based rendering.