November 20 – 25 , 2011, Dagstuhl Seminar 11471

Efficient Algorithms for Global Optimisation Methods in Computer Vision


Andrés Bruhn (Universität des Saarlandes, DE)
Thomas Pock (TU Graz, AT)
Xue-Cheng Tai (University of Bergen, NO)

For support, please contact

Dagstuhl Service Team


Dagstuhl Report, Volume 1, Issue 11 Dagstuhl Report
List of Participants
Dagstuhl's Impact: Documents available
Dagstuhl Seminar Schedule [pdf]


Most of the leading algorithms in computer vision are based on global optimisation methods. Such methods compute the solution of a given problem as minimiser of suitable cost functional that penalises deviations from previously made assumptions and integrates them over the entire image domain. While their transparent modelling allows for excellent results in terms of quality for many fundamental computer vision tasks such as motion estimation, stereo reconstruction, image restoration, shape matching, and object segmentation, the minimisation of these cost functional often leads to optimisation problems that are mathematically challenging and computationally expensive.

In the last decade, this fact has triggered a variety of different research directions that try to satisfy the needs of an ever increasing resolution in image and video data as well as the strong real-time demands of industrial applications. These research directions can be roughly divided into four fields that correspond to the different stages of the algorithmic design pipeline:

  • Modelling (suitable priors, continuous vs. discrete, hybrid approaches).
  • Mathematical analysis (convex vs. non-convex, error bounds, well-posedness).
  • Numerical Solvers (recent techniques, trends in numerical/combinatorial optimisation).
  • Parallelisation (GPUs, mutli-core, cluster systems, FPGAs).

Since there are no conferences that address all four fields, the goal of this seminar was to identify and address open questions in these four fields in the context of the entire design pipeline. To this end, it brought together computer scientists and mathematicians from all stages. Apart from stimulating interdisciplinary research and establishing close collaborations between the different fields by scheduling plenty of time for discussions, the ultimate goal of the seminar was to develop more precise and more efficient algorithms that are conceptually well designed, mathematically well understood and from which all parts are chosen carefully such that they harmonise with each other.

Further aims of the seminar were to establish suitable benchmarks for measuring the performance of each of the stages (model precision, optimisation accuracy, numerical efficiency, parallelisability) and to derive general conceptual guidelines for the design of efficient algorithms that are applicable to a broad class of key problems in computer vision.

The seminar was conducted in a conference style, where every participant gave a talk of about 20 to 25 minutes. There was much time for extensive discussions - directly after the talks, in dedicated working groups and in the evenings. As documented by the very positive evaluation and the detailed summaries of the working groups, there was a very open and constructive atmosphere. In particular the people appreciated the integration of young researchers in the seminar schedule and the gain of new cross-disciplinary insights from talks and discussions of other participants. While the first observation reflects the fact that the field of efficient algorithms in computer vision is relatively new, the second aspect demonstrates the success of the seminar to bring together people from different communities and establish new ties between the fields. At the moment that this report is written it is very difficult to identify new fundamental issues of efficient algorithms in computer vision. However, lots of interesting aspects were discussed during the talks and in the discussion groups, and many participants established collaborations with people from other fields. Thus we believe this seminar served as an excellent basis to inspire and trigger novel developments in the design of efficient algorithms for global optimisation problems in computer vision.

Finally, it should be mentioned that there was a huge consensus among the participants that there should be a follow-up event to this seminar in the upcoming years. For the current seminar, there will be edited post-proceedings in the Springer LNCS series which gives all participants the opportunity to summarise results from the seminar, discuss open questions and present recent research in the field. The deadline for submission is end of March 2012.


  • Computer Vision
  • Modelling
  • Data Structures / Algorithms
  • Other Category: Mathematical Foundations
  • Parallel Computing


  • Continuous and discrete optimisation methods
  • Variational methods
  • Partial differential equations
  • Primaldual algorithms
  • Hierarchical strategies
  • Multigrid methods
  • Parallel computing
  • GPU-based implementations


In the series Dagstuhl Reports each Dagstuhl Seminar and Dagstuhl Perspectives Workshop is documented. The seminar organizers, in cooperation with the collector, prepare a report that includes contributions from the participants' talks together with a summary of the seminar.


Download overview leaflet (PDF).


Furthermore, a comprehensive peer-reviewed collection of research papers can be published in the series Dagstuhl Follow-Ups.

Dagstuhl's Impact

Please inform us when a publication was published as a result from your seminar. These publications are listed in the category Dagstuhl's Impact and are presented on a special shelf on the ground floor of the library.