Welcome to the homepage of the

Mathematical Optimization Group

Department of Mathematics, University of Tübingen, Germany

An iterated L1 Algorithm for Non-smooth Non-convex Optimization in Computer Vision

P. Ochs, A. Dosovitskiy, T. Pock and T. Brox

Abstract:
Natural image statistics indicate that we should use nonconvex norms for most regularization tasks in image processing and computer vision. Still, they are rarely used in practice due to the challenge to optimize them. Recently, iteratively reweighed l1 minimization has been proposed as a way to tackle a class of non-convex functions by solving a sequence of convex l2-l1 problems. Here we extend the problem class to linearly constrained optimization of a Lipschitz continuous function, which is the sum of a convex function and a function being concave and increasing on the non-negative orthant (possibly non-convex and nonconcave on the whole space). This allows to apply the algorithm to many computer vision tasks. We show the effect of non-convex regularizers on image denoising, deconvolution, optical flow, and depth map fusion. Non-convexity is particularly interesting in combination with total generalized variation and learned image priors. Efficient optimization is made possible by some important properties that are shown to hold.
pdf Bibtex Publisher's link Code
Citation:
P. Ochs, A. Dosovitskiy, T. Pock, T. Brox:
An iterated L1 Algorithm for Non-smooth Non-convex Optimization in Computer Vision. [pdf]
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2013.
Bibtex:
@inproceedings{ODBP13,
  title        = {An iterated L1 Algorithm for Non-smooth Non-convex Optimization in Computer Vision},
  author       = {P. Ochs and A. Dosovitskiy and T. Pock and T. Brox},
  year         = {2013},
  booktitle    = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
}


MOP Group
©2017-2022
The author is not
responsible for
the content of
external pages.