To view the poster program please click here

Back to overview


A new Image Segmentation Tool powered by Machine Learning

Thursday (27.09.2018)
11:45 - 12:00 S1/01 - A04
Part of:

When analyzing the microstructure of materials it is no longer sufficient to just acquire images. Automated image analysis becomes more and more important to extract meaningful information and generate understanding. Unfortunately, available analysis techniques often require already segmented input data which is not readily available. Image segmentation is not straight-forward and can become a tedious manual task in case of microscopic images that are prone to artefacts and noise. In this contribution we present a new software module that provides easy-to-use image segmentation using machine learning techniques.


Thresholding and watershed algorithms are probably the most common tools for image segmentation, but their success strongly depends on the quality of the images. They only perform well on clean and noise-free data, whereas machine learning based approaches have shown huge potential to tolerate a certain amount of variance in the input data. However, currently available solutions are not easy to use and often afford image analysis expert skills to apply them successfully. Therefore the presented software has been developed with ease of use in mind. It is seamlessly integrated in the microscope’s acquisition and analysis software for direct access. It features an intuitive user interface for training the model. No parameters have to be set or known. The training process relies on labeling with a simple painting tool and provides immediate feedback about the segmentation result. This way the model can easily be refined and retrained. Once it is ready it can be applied to segment similar data repeatedly.


The presented solution works with any image data covering 2D and 3D datasets from light-, electron-, ion- or x-ray-microscopes and can even deal with 6D datasets by high-end light microscopy. Users can be trained within minutes to operate the software module and perform advanced image segmentation. This enables user-independent and reproducible segmentation for everybody and facilitates consistent automated image analysis. [1]




[1] Visit for further information and your free 30-day trial version.

Tobias Volkenandt
Carl Zeiss Microscopy GmbH
Additional Authors:
  • Stefanie Freitag
    Carl Zeiss Microscopy GmbH
  • Dr. Michael Rauscher
    Carl Zeiss Microscopy GmbH