CALL FOR PAPERS
Special Issue of the International Journal of Computer Vision on
PROBABILISTIC MODELS FOR IMAGE UNDERSTANDING

Aims and scope: Probabilistic models provide a compelling framework for
describing image and video content at levels ranging from small image
patches to overall scene and motion structure. We solicit papers
describing the development, learning and use of principled probabilistic
models for image understanding. Relevant topics include (but are not
limited to):
- low level models (image patches, random fields),
- object recognition / detection,
- structural models / image parsing,
- structured models of human motion,
- probabilistic frameworks for image representation,
- efficient algorithms for learning such models,
- frameworks and datasets for evaluating such models.

We are particularly interested in models:
- that incorporate rich structure (deep, graphical, hierarchical,
compositional,...) or that are suited for use within structure-based
frameworks,
- or that minimize the amount of labelled data required for learning
new classes, by exploiting latent structure or reusing components
or priors.

We will consider submissions describing specific models, position
papers and evaluation papers:
- Papers on specific models should either give enough detail for a
moderately skilled graduate student to reimplement the exact model
used and reproduce the experimental results, or provide a full
description or an open source implementation as supplementary
material. They must also include a discussion of competing
approaches and comparative testing that establishes the advantages
and limitations of the approach presented.
- Position papers should clearly present the arguments for and against
one or more generic approaches, supporting these with indicative
experimental results, comparative tables and thorough discussions of
the existing literature.
- Evaluation papers should describe a benchmark or challenge problem,
motivating it by discussing limitations of existing models or
benchmarks or debates regarding performance that need to be
resolved, presenting the detailed evaluation methodology and the
dataset (coverage, collection, labelling), and presenting a
representative sample of benchmark results for baseline methods or
recent methods from the literature. The benchmark and dataset must
be non-proprietary and publicly available so that other authors can
test their own methods on it. If possible open source
implementations of the baseline models and feature sets should also
be made available.

Submissions: Papers following the usual IJCV author guidelines should
be submitted to the IJCV website http://visi.edmgr.com , choosing the
Special Issue article type "Probabilistic Models for Image
Understanding". Regular journal articles (25 pages) are preferred,
but short papers (10 pages) and well-balanced surveys (30 pages) will
also be considered. All submissions will be subject to peer review.

Submissions will be returned without review if we feel that they are not
well aligned with the goals of the special issue. If you are unsure
whether an intended submission is in scope, send an abstract or a draft
to the editors of the special issue at least one month before the
submission deadline.

Submission deadline: July 21 2008
Scheduled publication date: Summer 2009

Guest editors:
- Bill Triggs, mailto:Bill.Triggs@imag.fr
- Chris Williams, mailto:ckiw@inf.ed.ac.uk