EPFL
 Biomedical Imaging Group
EPFL   Student Projects: Daniel Stadelmann
English only    BIG > Teaching > Student Projects > Daniel Stadelmann
 CONTENTS
Home page
Events
Members
Publications
Tutorials and Reviews
Research
Demos
Download Algorithms
Teaching
bulletStudent projects

Watershed by Flooding


Daniel Stadelmann

Micro-engineering, EPFL


Semester project

June 2002


Interactive demo

Applet demo


Motivation

Robotics
In plants producing praline products a Delta robot picks up the chocolates from a conveyor belt. Then they are put into a box. A vision system tracks the chocolates and guides the robots.
Sometimes it occurs that two chocolates on the belt touch each other. The difficulty is to separate them. The robot actually has to realize that it is dealing with two objects and has to grip them separately.

Biological and Medical Applications
There is an immense interest by biologists in segmenting images. For example cells are often quite close to each other if not even touching.
Searchers investigate on the migration of cells and want to track or count them. Image processing and especially morphological image analysis is a promising technique to investigate probes. To achieve this, however, segmenting the image is often a prerequisite.


Flooding the Image

This implementation of the watershed algorithm is based on the immersion analogy. A gray level image is seen as a topological map, the gray values representing the elevation.
The image is then flooded by water sources situated at every local minimum. As soon as the water coming from one source touches the water coming from a neighboring source, a dam is built. This dam represents a watershed line which separates adjacent objects.

Figure 1:Two objects separated by a watershed. Figure 2:Profile of figure 1 showing the two sources and the dam ("watershed") between them.

Results

Figure 3:The red lines represent the watershed lines, separating neighboring cells. Figure 4:Some of the sand grains are over segmented due to several local minima on one object.