Gianfranco Doretto / Publications

Dynamic textures

Doretto, G., Pundir, P., Wu, Y. N., and Soatto, S.
Dynamic textures
Technical Report TR200032, UCLA Computer Science Department, 2000.

Download

PDF (586.6kB )  

Abstract

Dynamic textures are sequences of images of moving scenes that exhibit certain stationariety properties in space and time; these include sea-waves, smoke, foliage, whirlwinds etc. We present a novel characterization of dynamic texture that poses the problems of modeling, learning, recognizing and synthesizing dynamic textures on a firm analytical footing. We borrow tools from stochastic realization theory to capture the “essence” of dynamic textures; we do so by learning models that are optimal in the sense of maximum likelihood or minimum prediction error variance. Once learned, therefore, a model has predictive power and can be used for extrapolating synthetic sequences to infinite length with negligible computational cost. We present experimental evidence that, within our framework, even low dimensional models can capture very complex visual phenomena.

BibTeX

@TechReport{UCLA-CSD-TR200032,
  Title                    = {Dynamic textures},
  Author                   = {Doretto, G. and Pundir, P. and Wu, Y. N. and Soatto, S.},
  Institution              = {UCLA Computer Science Department},
  Year                     = {2000},
  Address                  = {Los Angeles, CA, USA},
  Month                    = dec,
  Number                   = {TR200032},
  Abstract                 = {Dynamic textures are sequences of images of moving scenes that exhibit certain stationariety properties in space and time; these include sea-waves, smoke, foliage, whirlwinds etc. We present a novel characterization of dynamic texture that poses the problems of modeling, learning, recognizing and synthesizing dynamic textures on a firm analytical footing. We borrow tools from stochastic realization theory to capture the “essence” of dynamic textures; we do so by learning models that are optimal in the sense of maximum likelihood or minimum prediction error variance. Once learned, therefore, a model has predictive power and can be used for extrapolating synthetic sequences to infinite length with negligible computational cost. We present experimental evidence that, within our framework, even low dimensional models can capture very complex visual phenomena.},
  Bib2html_pubtype         = {Tech Reports},
  Bib2html_rescat          = {Dynamic Textures, Visual Motion Analysis, Visual Motion Recognition},
  File                     = {UCLA-CSD-TR200032.pdf:doretto\\report\\UCLA-CSD-TR200032.pdf:PDF;UCLA-CSD-TR200032.pdf:doretto\\report\\UCLA-CSD-TR200032.pdf:PDF},
  Keywords                 = {textures, dynamic scene analysis, 3D textures, minimum description length, image priors, image compression, generative model, prediction error methods, ARMA regression, system identification, learning, E-M algorithm},
  Owner                    = {doretto}
}