Learning image segmentation with waterpixels

Waterpixels on the L'Oréal melanocytes database

Part of PhD project, under the direction of Etienne Decencière and Thomas Walter.
In collaboration with L'Oréal Recherche et Innovation, France.

Waterpixels can be a pertinent tool for image segmentation learning, e.g. when the latter is seen as a classification task. Whereas superpixels are often used in the literature as primitives in such pipeline, we propose in this work to use them as support to compute new features for pixel classification. We call such features "SAF" (Superpixel-Adaptive Features).


SAF in a nutshell

One common approach in the pixel classification pipeline is to compute features on sliding windows centered on each pixel. In this work, we propose to use waterpixels, instead of windows, as computational support, in order to improve segmentation performance.

Sliding windows and superpixels (SAF) based features for pixel classification.

Back to research page