Scale-adaptive forest training via an efficient feature sampling scheme
Peter L, Pauly O, Chatelain P, Mateus D, Navab N (2015)
Publication Type: Book chapter / Article in edited volumes
Publication year: 2015
Journal
Publisher: Springer Verlag
Series: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Book Volume: 9349
Pages Range: 637-644
DOI: 10.1007/978-3-319-24553-9_78
Abstract
In the context of forest-based segmentation of medical data, modeling the visual appearance around a voxel requires the choice of the scale at which contextual information is extracted, which is of crucial importance for the final segmentation performance. Building on Haar-like visual features, we introduce a simple yet effective modification of the forest training which automatically infers the most informative scale at each stage of the procedure. Instead of the standard uniform sampling during node split optimization, our approach draws candidate features sequentially in a fine-to-coarse fashion. While being very easy to implement, this alternative is free of additional parameters, has the same computational cost as a standard training and shows consistent improvements on three medical segmentation datasets with very different properties.
Involved external institutions
Germany (DE)How to cite
APA:
Peter, L., Pauly, O., Chatelain, P., Mateus, D., & Navab, N. (2015). Scale-adaptive forest training via an efficient feature sampling scheme. In (pp. 637-644). Springer Verlag.
MLA:
Peter, Loic, et al. "Scale-adaptive forest training via an efficient feature sampling scheme." Springer Verlag, 2015. 637-644.
BibTeX: Download