The　observation　model　plays　a　key　role　in　performance　improvement　of　the　super　resolution　algorithms.　The　author　proposed　a　weight-matrix　based　blind　super　resolution　algorithm:　a　new　observation　model　based　on　a　motion　compensate　matrix　and　a　weight-matrix　is　defined　first,　then　the　high　resolution　image　and　the　weight　matrix　were　joint　estimated　by　alter　minimization　method　under　the　framework　of　Maximum　A　Prior　(MAP).　Evaluated　by　both　still　and　active　image　sequences,　the　algorithm　can　descript　the　degrading　process　of　the　observed　low　resolution　images　more　accurately.　It　shows　obvious　performance　improvement　compared　with　the　traditional　super　resolution　algorithm.　For　several　cases,　it　has　even　exceeded　the　results　when　the　observation　model　is　known.