Bayesian models for the emergence of visually moving objects

Abstract

Event Abstract Back to Event Bayesian models for the emergence of visually moving objects Rudolf Mester1, 2* and Christian Conrad1 1 Goethe Universität Frankfurt, FB12, Germany 2 Linköping University, ISY, Sweden The analysis of unconstrained, realistic, everyday dynamic visual stimuli and the interpretation of what we see as a temporally changing arrangement of objects in our environment is one of the central challenges in computational vision science. We report on recent insights obtained in investigations that aim at this ambitious (far) goal, experimenting with motion analysis and segmentation of video streams. In the present work, we emphasize a particular interpretation of the general Bayesian approach. The term 'Bayesian', in this context, means that prior knowledge (experience, learned statistics) is actively used in the process of determining what that state of the 'world outside' is, given the current (visual) observations, and all interpretations accumulated from the past. The required models in which a formulation of observation and estimated state of the outside world is performed can have different degrees of agreement with what is known from a physically realistic description of the process: A simplified, though still very useful model formulates the interpretation of visual input in terms of 'regions' which a) move according to naive physical laws, b) have a shape that usually only changes slowly, and c) show a surface 'texture' that is also roughly constant over time. On the basis of such a model, time-recursive estimation procedures can be developed which, on one hand, provide a 'dynamic segmentation' of a moving visual scene into regions (=the projections of 3D objects), and also allows to make predictions on the future visual observations. The actual implementation of such a procedure is impeded by the mathematical structure of the underlying optimization problem: it has extremely many variates, and due to the nonlinear structure of the problem, it has local minima 'almost everywhere'. We propose decision-directed approaches for solving this very fundamental kind of problems, deliberately sacrificing guaranteed optimality against feasibility, where the term 'feasible' is meant both in a technical sense (implementable and showing reasonable run-time) as well as in a biological sense (= implementations on a neural substrate appear plausible). We show results of fundamental investigations with a hierarchical, dynamical segmentation scheme, as well as first experiments with a Bayesian decision-directed scheme. Acknowledgements This work was in parts funded by the German Federal Ministry of Education and Research (BMBF) in the Bernstein Fokus Neurotechnologie – Frankfurt Vision Initiative 01GQ0841, and supported by the ELLIIT programme, the Strategic Area for ICT research, funded by the Swedish Government. References Alvaro Guevara, Christian Conrad, and Rudolf Mester: Curvature oriented clustering of sparse motion vector fields. In Proc. IEEE Southwest Symposium on Image Analysis & Interpretation (SSIAI) Santa Fe, USA, April 2012 Rudolf Mester: A Bayesian view on matching and motion estimation. In Proc. IEEE Southwest Symposium on Image Analysis & Interpretation (SSIAI) Santa Fe, USA, April 2012 R. Mester, C. Conrad, and A. Guevara: Multichannel Segmentation Using CR: Fast Super-Pixels and Temporal Propagation. Scandinavian Conference on Image Analysis, SCIA 2011, Ystad, May 2011 (Springer LNCS series, vol. 6688) Keywords: Bayesian Analysis, object grouping, segmentation, Visual Motion Processing, Visual Perception Conference: Bernstein Conference 2012, Munich, Germany, 12 Sep - 14 Sep, 2012. Presentation Type: Poster Topic: Data analysis, machine learning, neuroinformatics Citation: Mester R and Conrad C (2012). Bayesian models for the emergence of visually moving objects. Front. Comput. Neurosci. Conference Abstract: Bernstein Conference 2012. doi: 10.3389/conf.fncom.2012.55.00108 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 17 May 2012; Published Online: 12 Sep 2012. * Correspondence: Prof. Rudolf Mester, Goethe Universität Frankfurt, FB12, Frankfurt, Germany, mester@vsi.cs.uni-frankfurt.de Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Rudolf Mester Christian Conrad Google Rudolf Mester Christian Conrad Google Scholar Rudolf Mester Christian Conrad PubMed Rudolf Mester Christian Conrad Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.