High Entropy Ensembles for Holistic Figure-ground Segmentation
In Proceedings British Machine Vision Conference 2014
http://dx.doi.org/10.5244/C.28.105
Abstract
Modern computer vision applications are built on processing pipelines with hard and intensive tasks; among those, figure-ground segmentation is definitely one of the most important and challenging. As proved by many works in literature, an effective way of approaching this task consists in combining different algorithms in structures where they synergically collaborate towards the solution of the given segmentation problems. Inspired by other model combination frameworks, we propose a novel method to create graph-based ensembles of randomly configured figure-ground segmentation algorithms. The graph-based topology enables the algorithms to communicate with each other to let the strengths of one overcome the weaknesses of the others and vice versa, while the randomness injection reduces the risk of overfitting, decreases the computational complexity of the model creation procedure and enables our ensembles to overcome state-of-the-art results for several challenging datasets.
Session
Poster Session
Files
Extended Abstract (PDF, 1 page, 177K)Paper (PDF, 13 pages, 7.2M)
Bibtex File
Citation
Ignazio Gallo, Alessandro Zamberletti, Simone Albertini, and Lucia Noce. High Entropy Ensembles for Holistic Figure-ground Segmentation. Proceedings of the British Machine Vision Conference. BMVA Press, September 2014.
BibTex
@inproceedings{BMVC.28.105
title = {High Entropy Ensembles for Holistic Figure-ground Segmentation},
author = {Gallo, Ignazio and Zamberletti, Alessandro and Albertini, Simone and Noce, Lucia},
year = {2014},
booktitle = {Proceedings of the British Machine Vision Conference},
publisher = {BMVA Press},
editors = {Valstar, Michel and French, Andrew and Pridmore, Tony}
doi = { http://dx.doi.org/10.5244/C.28.105 }
}