Bayesian object localisation in images

Citation
J. Sullivan et al., Bayesian object localisation in images, INT J COM V, 44(2), 2001, pp. 111-135
Citations number
46
Categorie Soggetti
AI Robotics and Automatic Control
Journal title
INTERNATIONAL JOURNAL OF COMPUTER VISION
ISSN journal
09205691 → ACNP
Volume
44
Issue
2
Year of publication
2001
Pages
111 - 135
Database
ISI
SICI code
0920-5691(200109)44:2<111:BOLII>2.0.ZU;2-X
Abstract
A Bayesian approach to intensity-based object localisation is presented tha t employs a teamed probabilistic model of image filter-bank output, applied via Monte Carlo methods, to escape the inefficiency of exhaustive search. An adequate probabilistic account of image data requires intensities both i n the foreground (i.e, over the object), and in the background, to be model led. Some previous approaches to object localisation by Monte Carlo methods have used models which, we claim, do not fully address the issue of the st atistical independence of image intensities. It is addressed here by applyi ng to each image a bank of filters whose outputs are approximately statisti cally independent. Distributions of the responses of individual filters, ov er foreground and background, are learned from training data. These distrib utions are then used to define a joint distribution for the output of the f ilter bank, conditioned on object configuration, and this serves as an obse rvation likelihood for use in probabilistic inference about localisation. The effectiveness of probabilistic object localisation in image clutter, us ing Bayesian Localisation, is illustrated. Because it is a Monte Carlo meth od, it produces not simply a single estimate of object configuration, but a n entire sample from the posterior distribution for the configuration. This makes sequential inference of configuration possible. Two examples are ill ustrated here: coarse to fine scale inference, and propagation of configura tion estimates over time, in image sequences.