A texton-based cloud detection algorithm for MSG-SEVIRI multispectral images

Abstract

A new statistical texton-based method for cloud detection through satellite image analysis is presented. The ultimate goal is to improve the performance of remote sensing techniques used to support the observations of active volcanic processes. The proposed method is a supervised classifier that exploits radiance spatial correlation in satellite images using a statistical descriptor of texture called texton. Cloudy and clear-sky models are determined using cluster analysis over the image features. The pixels to be classified are compared with the estimated models and assigned to the closest model. The cloud detection algorithm has been tested on a data set of MSG-SEVIRI images acquired during 2008 (about 35,000 images) of the Sicily area. Results show that the texton-based approach is robust in terms of percentage of correctly classified pixels, reaching more than 85% of success in both daytime and nighttime images. © 2011 Taylor & Francis.