Dense random fields
Abstract/Contents
- Abstract
- Random field models are commonly used to express prior knowledge about image formation and are one of the central tools of computer vision and image processing. Prior random field algorithms were restricted to sparse graph structures, such as grids in which only adjacent image pixels are connected and exchange information. In this thesis, we introduce dense random fields (DRFs), in which all image pixels are connected directly. DRFs encode a much richer prior, capturing interactions between distant elements in an image. Dense random fields have billions of edges, making traditional algorithms impractical. I will show that a representation of pairwise connections by combinations of Gaussian kernels enables highly efficient algorithms for inference and learning in DRFs. Our main inference algorithm, based on the mean field approximation, converges in a fraction of a second and yields state-of-the-art results on semantic image segmentation. This algorithm forms the basis for a family of learning algorithms that further improve DRF performance. Finally, we present an inference algorithm for continuous DRFs that can be used for dense motion estimation.
Description
Type of resource | text |
---|---|
Form | electronic; electronic resource; remote |
Extent | 1 online resource. |
Publication date | 2014 |
Issuance | monographic |
Language | English |
Creators/Contributors
Associated with | Kräehenbüehl, Philipp |
---|---|
Associated with | Stanford University, Department of Computer Science. |
Primary advisor | Koltun, Vladlen, 1980- |
Thesis advisor | Koltun, Vladlen, 1980- |
Thesis advisor | Li, Fei Fei, 1976- |
Thesis advisor | Savarese, Silvio |
Advisor | Li, Fei Fei, 1976- |
Advisor | Savarese, Silvio |
Subjects
Genre | Theses |
---|
Bibliographic information
Statement of responsibility | Philipp Kräehenbüehl. |
---|---|
Note | Submitted to the Department of Computer Science. |
Thesis | Thesis (Ph.D.)--Stanford University, 2014. |
Location | electronic resource |
Access conditions
- Copyright
- © 2014 by Philipp Kraehenbuehl
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...