Image complexity measure: a human criterion free approach

This paper describes a novel fuzzy approach to determine the complexity of an image. The proposed method determines image complexity based on analysis of edge level percentages in the image. The method determines first the complexity class of the image, little complex, more or less complex, and very complex using centroids obtained from a fuzzy clustering process. Then the membership to that class is computed by a set of interval mapping functions. The method is very consistent because it does not incorporate any a priori human evaluation of complexity. Results of the method show correlation with human complexity values obtained in an evaluation test, however the values obtained with the method are more consistent. The paper also shows satisfactory results of the method in an application on an image edge detection task.