Extraction of Diffuse Objects From Images by Means of Watershed and Region Merging: Example of Solar Images

Abstract

This paper presents a new method for extraction of diffuse objects from images, which was developed for segmentation of solar images obtained from extreme-UV imaging telescope (EIT) experiments of the satellite SOHO mission. As a particular type of objects to be extracted coronal holes in EIT images have been chosen. The method described is based on the use of a watershed algorithm. The result of the watershed segmentation is a partition of the whole domain of the image into a large number of small regions. These regions are then combined in a region merging process. The proposed region merging algorithm iteratively adds the darkest regions and maximizes the average contrast between a current mask and a set of its neighboring regions. We show that the maximization of the average contrast gives segmentation results that are visually acceptable. Furthermore, this approach allows us to conduct the segmentation of EIT images independently of any explicit fine-tuning parameters. The proposed method was extensively tested on EIT images obtained at various times and various levels of solar activity, and we will show that it can be used independently of the local brightness level and the extent of coronal holes.

Automated coronal hole segmentation from Solar EUV Images using the watershed transform

“…It should be noted that compared to the method proposed in [10], the operating scheme of this method excludes image region labelling, which in this study is an element of the segmentation by watershed immersion, characterised by a linear time complexity [34,35]. In addition, this project made use of the algorithm for lower completion [35], which gives the segmentation by watershed immersion watershed lines 1 pixel wide, thus providing equal division of pixels between different regions.…”

“…It should be noted that, after the ring is superposed on the solar image, the operations presented in Figure 3 are executed inside the area of the ring. The novel solutions proposed measurably result in a significantly shorter operating time of the segmentation method compared to [10]. Individual steps of the method illustrated in Figure 3 will be presented in subsections below.…”

“…On the other hand, region-based methods are used to detect sunspots [6,7], thread-like filaments [8,9], coronal holes [10,11], and active regions [12]. Region-based approaches use statistical data from sub-regions and enable the optimum energy based on which the model fits the image best to be found.…”

“…In this study, the watershed segmentation method [34,35,36] first presented in [10] was used to identify areas of coronal holes. The watershed segmentation method makes use of the notions from topography, but expressed in the mathematical morphology language [34].…”

“…This study introduces new solutions into the process of segmenting coronal holes from EIT195 images compared to [10]. They relate both to the watershed segmentation algorithm used and the method of merging segments after the segmentation into the resultant coronal hole area.…”