Most　previous　super-resolution　(SR)　approaches　are　implemented　with　two　individual　cascade　steps,　image　registration　and　image　fusion,　which　handicaps　the　incorporation　of　the　structural　information　of　the　objects　of　interest,　e.g.　human　faces,　into　SR　in　a　parallel　way.　This　prior　information　is　beneficial　to　either　robust　motion　estimation　or　fusion　with　higher　quality.　In　this　paper,　SR　reconstruction　is　formulated　as　Bayesian　state　estimation　of　location　and　appearance　parameters　of　a　face　model.　In　addition,　a　sequential　Monte　Carlo　(SMC)　based　algorithm　is　proposed　to　achieve　the　probabilistic　state　estimation,　i.e.　SR　reconstruction　in　our　formulation.　Image　alignment　and　image　fusion　are　combined　into　one　unified　framework　in　the　proposed　approach,　in　which　the　prior　information　from　the　face　model　is　incorporated　into　both　registration　and　fusion　process　of　SR.　Experiments　performed　on　synthesized　frontal　face　sequences　show　that　the　proposed　approach　gains　superior　performance　in　registration　as　well　as　reconstruction.