Video Objects Segmentation Using Eulerian Region-Based Active Contours

We addressthe problemof moving objectsegmentation using active contours. As far as segmentationof moving objectsis concerned,region-basedtermsmustbe incorporatedin theevolutionequationof theactivecontour, in addition to classicalboundary-basedterms.In this paper, we proposea general framework for region-basedactivecontours. Novel aspectsof the segmentationmethodinclude a new Eulerianproof to computetheevolutionequationof theactivecontourfromtheminimizationof a criterion, and theintroductionof functionsnamed“descriptors” of theregions.In this proof, thedynamicalschemeis directlyintroducedin the criterion before differentiation. With such a method,thecaseof descriptors dependingon theevolution of thecurve, i.e. dependinguponfeaturesgloballyattached to the region, canreadilybe taken into account.Thevariation of thesedescriptors upon the evolution of the curve inducesadditional termsin the evolution equationof the activecontour. Theproofensuresthefastestdecreaseof the activecontourtowardsa minimumof the criterion. Inside this theoretical framework,a setof descriptors is evaluated onreal sequences for thedetectionof moving objects.