Based　on　a　novel　two-dimensional　autoregressive　moving　average　(2D-ARMA)　parameter　estimate,　this　paper　develops　a　neural　network　algorithm　for　fast　blind　image　restoration.　The　point　spread　function　of　degraded　image　is　reformulated　as　an　optimal　solution　of　a　quadratic　convex　programming　problem　and　it　is　well　solved　by　a　neural　network.　Compared　with　existing　ARMA　parametric　methods,　the　proposed　approach　can　overcome　the　local　minimization　problem.　Unlike　iterative　blind　deconvolution　algorithms,　the　proposed　blind　image　restoration　algorithm　has　a　faster　blind　image　restoration.　Computed　results　shows　that　the　proposed　algorithm　can　obtain　a　better　image　estimate　with　a　faster　speed　than　two　standing　blind　image　restoration　algorithms.　©2010　IEEE.