This　paper　researches　the　mutual　information　matching　algorithm　for　image　registration.　Firstly,　feature　points　are　extracted　from　images　by　Harris　corner　detector　operator.　Then,　it　uses　mutual　information　as　the　similarity　criteria　for　feature　points＇　coarse　matching.　At　last,　it　adapts　random　sample　consensus　algorithm　(RANSAC)　to　filter　the　results　of　the　coarse　matching　in　order　to　eliminate　errors　and　inconsistent　with　the　camera　motion　matching　point.　We　discuss　the　window　size　and　the　number　of　histogram　bins　which　influence　the　result　of　registration.　Large　experimental　results　show　that　the　method　can　make　a　satisfactory　registration　accuracy　and　stability.　©　2011　IEEE.