In　this　paper,　we　propose　a　novel　method　for　solving　single-image　super-resolution　problems.　Firstly,　using　the　human　visual　perception　and　image　gradient　features,　the　image　total　variation　is　decomposed　into　structural　components　and　texture　components.　Based　on　the　theory　about　sparse　signal　representation,　we　used　K-SVD　method　to　generate　ultra-complete　dictionary　and　to　achieve　the　reconstruction　of　the　texture　component.　Then　the　super-resolution　reconstruction　of　the　whole　original　low　resolution　image　is　realized　by　fused　them　with　the　bi-cubic　interpolated　image　reconstruction　of　the　structural　components.　The　proposed　method,　without　external　image　database　support,　brings　in　the　whole　image　information　while　depends　on　the　fixed　-　K　neighborhood.　It　can　upgrade　the　fitting　performance　of　the　existing　methods,　and　enhance　mole　detail　image　information,　also　improve　the　reconstructed　image　quality.　©　2010　IEEE.