Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6937462 | Computer Vision and Image Understanding | 2017 | 22 Pages |
Abstract
Graph-cuts based move making algorithms have been intensively studied. Previous methods uniformly rely on max-flow/min-cut solutions for move-making, and have achieved generally good performance on a variety of applications. Early research suggests that path-augmenting algorithms such as BK tend to perform well on grid-structured graphs. Unlike conventional graph-cuts methods, our algorithm does not require the exact max-flow/min-cut solution for update. Instead, any cut/flow of a subproblem can be used for primal/dual update, which allows the max-flow solver to stop at any time during execution. Thanks to the dynamicity of our approach, the energy convergence rate can be improved by several times in our experiments on GPU.
Keywords
Related Topics
Physical Sciences and Engineering
Computer Science
Computer Vision and Pattern Recognition
Authors
Ziang Cheng, Yang Liu, Guojun Liu,