Due　to　limitations　on　the　image　capturing　devices,　distance,　storage　capability　and　bandwidth　for　transmission,　many　images　in　multimedia　applications　are　low-bit-rate　compressed　and　low　resolution.　In　this　paper,　we　proposed　a　class-specified　learning　based　super　resolution　for　this　kind　of　low　quality　images.　Firstly,　we　proposed　a　class-specified　filter　to　remove　the　compressed　distortions.　Then　a　class-specified　learning　based　scheme　is　employed　to　super　resolve　images　with　different　compression　rates.　Experimental　results　show　that　the　proposed　method　can　improve　both　the　objective　and　subjective　quality　of　the　images　effectively.　©　2014　IEEE.