Colour transformations and K-means segmentation for automatic cloud detection

Blazek, Martin; Pata, Petr

Meteorologische Zeitschrift Vol. 24 No. 5 (2015), p. 503 - 509

19 Literaturangaben

veröffentlicht: Aug 31, 2015
Online veröffentlicht: Aug 31, 2015
Manuskript akzeptiert: Jun 2, 2015
Manuskript-Revision erhalten: May 25, 2015
Manuskript-Revision angefordert: Feb 2, 2015
Manuskript erhalten: Oct 29, 2014

DOI: 10.1127/metz/2015/0656

BibTeX Datei


Open Access (Arbeit kann kostenlos heruntergeladen werden)

Arbeit kostenlos herunterladen


The main aim of this work is to find simple criteria for automatic recognition of several meteorological phenomena using optical digital sensors (e.g., Wide-Field cameras, automatic DSLR cameras or robotic telescopes). The output of those sensors is commonly represented in RGB channels containing information about both colour and luminosity even when normalised. Transformation into other colour spaces (e.g., CIE 1931 xyz, CIE L*a*b*, YCbCr) can separate colour from luminosity, which is especially useful in the image processing of automatic cloud boundary recognition. Different colour transformations provide different sectorization of cloudy images. Hence, the analysed meteorological phenomena (cloud types, clear sky) project differently into the colour diagrams of each international colour systems. In such diagrams, statistical tools can be applied in search of criteria which could determine clear sky from a covered one and possibly even perform a meteorological classification of cloud types. For the purpose of this work, a database of sky images (both clear and cloudy), with emphasis on a variety of different observation conditions (e.g., time, altitude, solar angle, etc.) was acquired. The effectiveness of several colour transformations for meteorological application is discussed and the representation of different clouds (or clear sky) in those colour systems is analysed. Utilisation of this algorithm would be useful in all-sky surveys, supplementary meteorological observations, solar cell effectiveness predictions or daytime astronomical solar observations.


CIELABCIE 1931YCbCrK-meanscloud detectionconsumer digital camerameteorologyRGBWILLIAM