Multiresolution particle filters in image processing

Recursively estimating the likelihood of a set of parameters, given a series of observations, is a common problem in signal processing. The particle filter is now a well-known alternative to the Kalman filter. It represents the likelihood as a set of samples with associated weights and so can approximate any distribution. It can be applied to problems where the process model and/or measurement model is non-linear. We apply the particle filter to the problem of estimating the structure of a scene from n views of that scene, by applying the particle filter across image resolutions