Patchy aurora image segmentation based on block threshold LBP

Abstract

The proportion of the aurora to sky is an important property for geosciences research. Before calculation, a crucial step is to segment the region of aurora light from the background. An automatic aurora image segmentation algorithm, based on block threshold local binary patterns (BTLBP), is proposed. In the training stage, LBP operator is applied to an all-sky image without aurora light, pixel by pixel, to get the reference feature vector of whole sky image. This image is then divided into the same size blocks and LBP operator is applied to each of them. In comparison with the reference feature vector, a threshold is found. In the segmentation stage, an image containing aurora is divided into blocks, whose features are compared with the threshold, aurora block is then detected. Simple as it is, online implementation on huge dataset is possible. The experiment showed that the proposed method is satisfying visually.