A local binary pattern classification approach for cloud types derived from all-sky imagers

Cloud classification from ground-based observations is a challenging task that attracts increasing attention, favoured by the development of all-sky imaging equipment. In this work, we propose a new method for cloud type classification from all-sky images. Appropriate versions of two descriptors, Regional Local Binary Pattern (R-LBP) and Four Patch-Local Binary Pattern (FP-LBP), are employed on the images in order to extract not only global but also local textural information from the observed cloud type patterns. In the classification stage, a linear Support Vector Machine (SVM) scheme and Linear Discriminant Analysis (LDA) classifiers are adopted for the assignment of the corresponding cloud type label. Our method is evaluated against two state-of-the-art methods and their datasets consisting of 5000 and 2500 images, respectively. According to the results, the proposed method outperforms the previous ones. Due to its promising results and the novelty of local pattern information of clouds, the proposed methodology could be considered as the basis for future studies aiming to overcome the basic disadvantage of all-sky imaging algorithms: to provide regional cloud type information instead of one cloud type for the whole sky.