Detection of Active Regions in Solar Images using Visual Attention

This paper deals with the problem of processing solar images using a visual saliency based approach. The system consists of two main parts: 1) a pre-processing part carried out by using an enhancement method that aims at highlighting the Sun in solar images and 2) a visual saliency based approach that detects active regions (events of interest) on the pre-processed images. Experimental results show that the proposed approach exhibits a precision index of about of 70% and thus it is, to some extent, suitable to allow detection of active regions, without human assistance, mainly in massive processing of solar images. However, the recall performance points out that at the current stage of development the method has room for improvements in detecting some active areas, as shown the F-score index that at presently is about 60%. © 2011 Springer-Verlag Berlin Heidelberg