As time goes on, we accumulate more experience with, rather than better algorithms for, image segmentation. Although the title of this paper intrigues readers with the word “novel,” it is nonetheless another ambitious implementation of a watershed algorithm.
Compared to its predecessors, more smoothing components are added, including morphological reconstruction, top/bottom hat transformation, and internal/external markers to reduce over-segmentation, which is often an issue with watershed-like algorithms. As the main contribution, this reduction is repeatedly shown in all four of the experiments included. However, by looking at the results, I am reminded that there is often no single correct answer to image segmentation; rather, the solution is dependent on the implied criteria. The idea of reducing the number of segmented components does not seem well suited for the images in figures 2, 3, and 4; for example, many objects of interest are missed/unlabeled due to the reduction.
This may be a good paper to read only if your segmentation result looks messy and broken. Otherwise, ignore it since no truly novel technique is presented.