Video object segmentation : driving active contours based on watershed boundaries
C. De Roover · Digital Access to Libraries (Université catholique de Louvain (UCL), l'Université de Namur (UNamur) and the Université Saint-Louis (USL-B)) · 2006
Object segmentation is nowadays a vital component of special effect creation. In order to add virtual object or a virtual background in a video sequence a mask needs to be constructed. This mask makes the delineation between the parts that must remain in the final sequence and the parts where we can add the special effects. Such a mask can easily be constructed by filming people in front of a blue or green screen. The main topic of this thesis is the conception of an automatic method to find out such a mask without any blue or green screen. We first suggest a segmentation framework based on the watershed transform. However, the mask boundaries generated with this method are not smooth over time. This results in a temporal flickering, which is very annoying from a visual point of view. To remove this flickering, we then use active contours. Watershed boundaries are then seen as magnets that attract active contours. Active contours have the particularity to be smooth both from the spatial and from the temporal point of view. The mask resulting from the combination of watersheds and active contours has therefore the advantages of both methodologies. It is spatially accurate thanks to the watersheds and temporally stable thanks to the active contours.