Recently,　based　on　a　generalized　least　absolute　deviation　(GLAD)　method,　a　cooperative　recurrent　neural　network　(CRNN)　algorithm　for　image　restoration　was　developed.　It　was　shown　that　the　CRNN　algorithm　can　obtain　an　optimal　image　estimate　under　the　non-optimal　regularization　parameter.　However,　the　CRNN　algorithm　has　a　slow　convergence　rate　due　to　its　continuous-time　feature.　For　real-time　applications　of　the　GLAD　method,　this　paper　proposes　a　discrete-time　algorithm　for　fast　image　restoration.　The　proposed　discrete-time　algorithm　is　shown　to　be　globally　convergent　to　the　optimal　image　estimate　under　a　fixed　computational　step　length.　Simulation　results　show　that　the　proposed　discrete-time　algorithm　has　a　faster　convergence　rate　than　the　CRNN　algorithm.　©2009　IEEE.