Aurora image classification based on LDA combining with saliency information

Abstract

There are different shapes of auroras in the sky around the arctic pole and the antarctic pole and there are different physical meaning and significance for different auroras. Therefore, the research on classification of aurora images has significant scientific value. In this paper, an aurora image classification method based on LDA with saliency information (SI-LDA) is proposed. First, the salience information of aurora images is used to generate visual dictionary which enhances the semantic information of aurora images. Next, the aurora images are represented by SI-LDA. Finally, SVM is applied to classify aurora images. Experimental results show that the proposed method achieves high performance over other algorithms available. ©Copyright 2013, Institute of Software, the Chinese Academy of Sciences.