Add noise to input image:

Choose class 1-4, then click into the image to set label:

Choose regularization strength $\lambda$:

Turn on to show result:

Show result

Try dragging & dropping your own images into the window!

View in Firefox for fastest speed.

About this model

This page demonstrates how to solve a variational multiclass labeling problem. By finding the minimizer $u$ of the relaxed energy $$\int_\Omega \langle u(x),s(x)\rangle \mathrm{d}x + \lambda \mathrm{TV}(u)$$ over all functions $u:\Omega \to \Delta_L$, where $L$ is the number of labels, one obtains an "almost binary" solution, which can be thresholded if necessary.

Authors: Benedikt Loewenhauser, Kevin Schober, Jan Lellmann (lellmann@mic.uni-luebeck.de)

Controls

Open settings panel on the top left.
Optionally: Add Gaussian noise.
For every label mark typical pixels by clicking into the image.
Select a regularization strength $\lambda$.
Click to show the result.
You may also try different input images (input in the top right).

References

J. Lellmann, C. Schnoerr: Continuous Multiclass Labeling Approaches and Algorithms. SIAM J. Imaging Sci., 4(4), 2011.

Embed

Copy the text below into your web page. To set the size, modify the width/height and w/h parameters: Done