Detection of low clouds in Meteosat IR night-time images based on a contextual spatio-temporal labeling approach

Abstract

This paper presents an original method to detect low clouds at night from Meteosat IR images. The exploit relevant motion-based measurements as well as information on the temperature and the thermal structure of the elements constituting the scene. A classification in three classes (low clouds, clear sky and other clouds) is embedded in a Bayesian estimation framework associated with Markov random field (MRF) models. An unsupervised learning scheme of thermal parameters associated to each class has been developed. We also propose a spatially-progressive minimization procedure of the energy function starting from reliably labelled pixels and based on the deterministic realization algorithm, ICM. Contextual information on the label field is taken into account. Experimental results are reported and compared with ground truth. They demonstrate the efficiency of the proposed approach.