Unsupervised Representation and Understanding of Multi-View Images

We study the structure of multi-view images and review the notion of plenoptic hypervolumes. We then present a segmentation algorithm based on the level set method to extract such hyper-volumes and show how this segmentation can be used to dis-occlude hidden objects and to facilitate the understanding of the monitored scene. Finally, we highlight preliminary results on classification and clustering of plenoptic hyper-volumes based on a multi-dimensional version of the SIFT algorithm.