Watershed lines suppression by waterfall marker improvement and line-neighbourhood analysis

Abstract

In this paper, we present a new self-sufficient image segmentation method based on watershed-lines neighbourhood. This work-study uses two morphological concepts, watershed and waterfall, for the development of a line-based catalogue concerning neighbourhood analysis. Innovation in waterfall hierarchical evolution is achieved with a marker-constraint implementation along the process, improving detail suppression. Watershed lines are then separated into multiple semi-lines for a further line suppression by semi-lines neighbourhood analysis. In many cases, unless a feature has a high thinness, different from image background (i.e., a thin road, in which case can be defined by a single watershed line), its representation is a set of watershed-connected lines that makes ambiguous feature-line detection for recognition of image structures. Applying watershed to the morphological gradient image, feature contouring is better characterized for waterfall application.