Model-Based Segmentation of Cloud Structures In Satellite Image Sequences

Abstract : Meteorological image sequences acquired with remote sensing devices contain huge amounts of available data featuring the temporal evolution of highly deformable and complex structures. In this study is presented a robust and effective model-based segmentation procedure for approximating highly deformable cloud structures. The whole image sequence may be viewed as a 3D data set, but the classical techniques for performing 3D segmentation are not meaningful here, because cloud structures are not shapes in the classical sense. The model-based segmentation presented here uses level-sets controlled by particle systems. A specific energy formulation is introduced, permitting the use of robust conjugate gradient techniques. The model is validated on a real meteorological image sequence. We outline the generalizations of the this particle system to perform image analysis tasks on truly 3D images.