@inproceedings{11859d77053a4dcdb915316e35af74ba,
title = "Silhouette-based variational methods for single view reconstruction",
abstract = "We explore the 3D reconstruction of objects from a single view within an interactive framework by using silhouette information. In order to deal with the highly ill-posed nature of the problem we propose two different reconstruction priors: a shape and a volume prior and cast them into a variational problem formulation. For both priors we show that the corresponding relaxed optimization problem is convex. This leads to unique solutions which are independent of initialization and which are either globally optimal (shape prior) or can be shown to lie within bounds from the optimal solution (volume prior). We analyze properties of the proposed priors with regard to the reconstruction results as well as their impact on the minimization problem. By employing an implicit volumetric representation our reconstructions enjoy complete topological freedom. Being parameter-based, our interactive reconstruction tool allows for intuitive and easy to use modeling of the reconstruction result.",
keywords = "Convex Optimization, Image-Based Modeling, Single View Reconstruction",
author = "Eno T{\"o}ppe and Oswald, {Martin R.} and Daniel Cremers and Carsten Rother",
year = "2011",
doi = "10.1007/978-3-642-24870-2_5",
language = "English",
isbn = "9783642248696",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "104--123",
booktitle = "Video Processing and Computational Video - International Seminar, Revised Papers",
note = "International Seminar on Video Processing and Computational Video ; Conference date: 10-10-2010 Through 15-10-2010",
}