ABSTRACT
This chapter aims to separate the connected ice floes, especially between those where the junction lines are almost invisible in the image. The distance transform-based watershed segmentation is applied, and a combination of concave detection and neighboring-region merging is proposed to automatically reduce the over-segmentation caused by the watershed. The gradient magnitude, which has high pixel values along object edges and low pixel values elsewhere, is natural to use in the watershed transform for grayscale image segmentation. This is because the main criterion of the segmentation is the homogeneity of the intensity values of the objects present in the image. The number of segmented regions depends on the regional minima found in the inverse distance map. Usually, there is more than one unique regional minimum for each object, and this will usually induce over-segmentation. Image segmentation by using the watershed transform is powerful for connected objects segmentation, especially for separation of connected objects with invisible boundaries.
