Abstract
Many interactive image segmentation techniques are based on semi-supervisedlearning. The user may label some pixels from each object and the SSL algorithmwill propagate the labels from the labeled to the unlabeled pixels, findingobject boundaries. This paper proposes a new SSL graph-based interactive imagesegmentation approach, using undirected and unweighted kNN graphs, from whichthe unlabeled nodes receive contributions from other nodes (either labeled orunlabeled). It is simpler than many other techniques, but it still achievessignificant classification accuracy in the image segmentation task. Computersimulations are performed using some real-world images, extracted from theMicrosoft GrabCut dataset. The segmentation results show the effectiveness ofthe proposed approach.