Demonstrator for Image Segmentation

Project by: Dmitrij Schlesinger, Norman Koch, Patric Röhm


What is image segmentation?
It is an important sub-area within digital image processing. In our case, segmentation means the division of the image into content-related regions, whereby the criteria according to which the division takes place are defined by the user.

What will be segmented in the images?
Examples are the distinction between foreground and background in an image (e.g. buildings or people in a landscape), between certain characteristic areas (forests, settlements, open spaces, water surfaces, etc.) in topographic maps or satellite images, or between "interesting" and "uninteresting" areas in microscope images (identification of biological cells).

How does the demonstrator work?
Automated digital image segmentation is usually performed using machine learning (ML) methods, with deep neural networksdemos (DNNs) being particularly suitable when the widest possible applicability (for many different problem classes) is required. The demonstrator for image segmentation developed at the ZIH of the TU Dresden allows users to segment their own images according to criteria they specify.

What procedure do I have to follow to use it?
The web application includes the following functionalities:
  • Registration of the user in the system with his email address
  • Upload of his pictures
  • Annotating the "interesting" and "uninteresting" areas in some training pictures with the mouse
  • Implementation of the learning process of the DNN
  • Segmentation of further images for which no annotation has been made yet
  • Downloading the results
  • Deleting images

What do I have to pay attention to?
  • The demonstrator serves exclusively scientific purposes.
  • Own pictures can only be seen by the user himself and by the administrators of the tool.
  • The demonstrator is available worldwide.

If I have any questions about it, where I find answers?




I have read, understood and accepted the license terms and conditions.