Characteristic Pattern Based on Mathematical Morphology

Abstract

Shape analysis is a very important issue in image analysis and computer vision. This paper describes a methodology using morphological operations to classify shapes by their decomposed components according to morphological structuring elements. A method called characteristic pattern is introduced to extract unique representations of an object shape among other shapes. Shape size analysis using mathematical morphology was introduced by Serra [1] where size criteria are discussed and geometrical properties of morphological processing on shapes are presented with morphological measurement. With size criteria, local area size parameters and global shape size distribution are both counted in Lebesque measure. Recent development of shape distribution can be found in [3,4,5,9]. In [5], pattern spectrum is used to describe the size distribution. Shape decomposition is another approach toward shape analysis, similar to a morphological skeletonization process. In [9], decomposition is completed by using a simplest object component (a disk) and analysis of an image is through a union of disks. The characteristic pattern concept is introduced in Section 2 of this paper to provide another approach toward shape analysis and matching. The basic concept is to derive a specific pattern associated with the object shape while other shapes within sample space do not possess this pattern orientation. We call this pattern orientation characteristic pattern. Characteristic pattern can be used to identify objects and classify shapes through decomposition — a parallel process using only local neighborhood pixel information.