A statistical classification method for hierarchical irregular objects

Abstract

This paper introduces a method for classifying structured visual objects that appear frequently in meteorological, medical and biological imagery. The focused objects are taken to be highly irregular and composed of subobjects in a hierarchical manner. The approach consists three principal steps. At first, hierarchical objects are detected in an segmented image. Secondly, shape descriptors are used to extract information of the contours of the objects. Finally, a global description for an object is obtained by applying statistical moments. As the goal is to classify natural objects, the most challenging task is to tolerate irregularity present in both spatial and hierarchical levels. Experiments with artificial images show that the method combines succesfully shape descriptors and object hierarchy.