AAAI Publications, Twenty-Third International FLAIRS Conference

Font Size: 
An Experimental Evaluation of Popular Image Parameters for Monochromatic Solar Image Categorization
Juan Banda, Rafal Angryk

Last modified: 2010-05-06


In this work, we perform experimental evaluation of different image parameters, from the perspective of large-scale contentbased discovery of solar phenomena in NASA images. Many of the recently published papers, often from different areas of application (e.g. medical, military images), describe different parameters for image recognition and state high quality of the generated results. Overwhelmed with the number of proposed image parameters, and concerned with their practical applicability to large-scale solar data analysis, we decided to perform our own comparative analysis of these parameters. Because of the lack of broadly accepted Solar Images benchmark data sets, in this paper we introduce our own benchmark created using solar images registered during the TRACE mission. Using the first solar domain-specific benchmark data set that contains multiple types of phenomena, we investigate the correlation of different image parameters and their importance from the perspective of distinguishing between different types of solar phenomena. In this paper, we present recommendations on the image parameters that perform adequately for the solar data.


Solar Image Categorization, Correlation, Feature Selection, Image Parameters

Full Text: PDF