Proximity Priors for Variational Semantic Segmentation and Recognition

TitleProximity Priors for Variational Semantic Segmentation and Recognition
Publication TypeConference Paper
Year of Publication2013
AuthorsBergbauer, J., Nieuwenhuis C., Souiai M., & Cremers D.
Other Numbers3624
Abstract

In this paper, we introduce the concept of proximity priors into semantic segmentation in order to discourage thepresence of certain object classes (such as ’sheep’ and’wolf ’) ’in the vicinity’ of each other. ’Vicinity’ encompasses spatial distance as well as specific spatial directionssimultaneously, e.g. ’plates’ are found directly above ’tables’, but do not fly over them. In this sense, our approachgeneralizes the co-occurrence prior by Ladickyet al. [3], which does not incorporate spatial information at all, andthe non-metric label distance prior by Strekalovskiyet al. [11], which only takes directly neighboring pixels intoaccount and often hallucinates ghost regions. We formulate a convex energy minimization problem with an exactrelaxation, which can be globally optimized. Results onthe MSRC benchmark show that the proposed approach reduces the number of mislabeled objects compared to previous co-occurrence approaches.

Acknowledgment

This work was partially funded by the Deutscher Akademischer Austausch Dienst (DAAD) through a postdoctoral fellowship.

URLhttp://www.icsi.berkeley.edu/pubs/vision/proximitypriors13.pdf
Bibliographic Notes

Proceedings of the Workshop on Graphical Models for Scene Understanding at the International Conference on Computer Vision 2013 (ICCV 2013), Sydney, Australia

Abbreviated Authors

J. Bergbauer, C. Nieuwenhuis, M. Souiai, and D. Cremers

ICSI Research Group

Vision

ICSI Publication Type

Article in conference proceedings