Thispaperaddressestheproblemofvideoobjectsegmentation,wheretheinitialobjectmaskisgiveninthefirstframeofaninputvideo.WeproposeanovelspatiotemporalMarkovRandomField(MRF)modeldefinedoverpixelstohandlethisproblem.UnlikeconventionalMRFmodels,thespatialdependenciesamongpixelsinourmodela