Image　feature　matching　is　an　integral　task　for　many　computer　vision　applications　such　as　object　tracking,　image　retrieval,　etc.　The　images　can　be　matched　no　matter　how　the　image　changes　owing　into　the　geometric　transformation　(such　as　rotation　and　translation),　illumination,　etc.　Also　due　to　the　successful　application　of　the　deep　learning　in　image　processing,　the　deep　learning　method　has　an　advantage　in　feature　extraction　of　images.　In　this　paper,　we　adopt　a　deep　Convolutional　neural　network　(CNN)　model,　which　attention　on　image　patch,　in　image　feature　points　matching.　CNN　obtains　the　feature　by　convolution　kernel　which　parameters　are　achieved　by　learning.　So　it　has　strong　ability　to　express　feature.　Compared　with　other　methods,　experimental　results　indicate　the　proposed　method　has　higher　accuracy　and　completed　efficiently.　©　2018　IEEE.