Image　registration　is　the　significant　part　in　some　image　processing　applications.　The　scale　invariant　feature　transform　(SIFT)　is　the　most　commonly　used　as　feature-based　image　registration　algorithm.　This　algorithm　will　produce　high　dimensional　feature　descriptors.　The　best　bin　first　(BBF)　is　a　very　efficient　algorithm　to　find　the　nearest　neighbor　from　a　large　number　of　high　dimensional　feature　descriptors.　But　this　algorithm　has　a　shortage:　it　may　not　obtain　the　correct　matching　feature　points　when　large　scale　exits　among　images.　In　this　paper,　we　overcome　the　shortage　through　improving　best　bin　first　search　algorithm;　this　approach　is　named　improved　BBF.　The　paper　improves　BBF　from　two　aspects.　First,　we　add　a　flag　in　data　structure　of　k-d　tree　node,　which　represents　whether　the　specified　feature　descriptor　need　search.　This　improving　makes　up　the　shortage　effectively.　Second,　we　build　the　k-d　tree　by　selecting　the　most　feature　descriptors.　This　improving　further　improves　the　searching　speed.　Through　large　experiments,　we　demonstrate　the　effectiveness　of　the　improved　BBF　search　algorithm　and　the　execution　speed　will　increase　above　10%.　©　2011　IEEE.