In　order　to　extract　the　features　of　the　image　more　accurately,　a　deep　belief　network　(DBN)　based　image　feature　extraction　method　is　proposed.　However,　when　the　deep　belief　network　extracts　the　features　of　the　image,　it　is　easy　to　ignore　the　local　texture　features　of　the　image.　Then　the　block　local　local　binary　mode　is　introduced　to　extract　the　local　texture　features　of　the　image.　At　the　same　time,　to　improve　the　slow　learning　speed　of　the　network,　the　initial　weight　of　the　network　is　improved.　Finally,　the　proposed　network　is　tested　on　the　ORL　image　dataset.　The　results　show　that　the　proposed　method　not　only　improves　the　recognition　accuracy　of　the　network,　but　also　accelerates　the　convergence　speed　of　the　network　to　some　extent.　©　2019　IEEE.