Global image feature extraction using slope pattern spectra

Abstract

Traditionally, granulometries are obtained using a series of openings or closings with convex structuring elements of increasing size. Granulometries constitute a useful tool for texture and image analysis since they are used to characterize size distributions and shapes [1], [2]. The granulometric analysis of an image results in a signature of the image with respect to the granulometry used which is referred to as a granulometric curve or pattern spectrum. Granulometric curves are used as feature vectors [3] for applications such as segmentation [4] and texture classification [5]. For example [6], granulometries based on openings with squares of increasing size as structuring elements, were used to extract dominant bean diameter from binary images of coffee beans. Granulometries were also used to estimate the dominant width of the white patterns in the X-ray images of welds [7]. Due to the computational load associated with the calculation of granulometries, Vincent [6], building on the work of Haralick et al. [2] and proposed fast and efficient granulometric techniques using linear openings.