Image　feature　matching　is　an　important　part　of　SLAM　(Simultaneous　Localization　and　Mapping　algorithm).　In　order　to　improve　the　implementation　efficiency　of　standard　RANSAC　algorithm,　this　paper　proposed　a　novel　improved　RANSAC　algorithm　to　deal　with　the　mismatch　in　the　image　matching　procedure.　Our　method　deals　with　raw　sample　data　and　predict　the　inliers　in　the　sample　data　according　to　the　Euclidean　distance　between　feature　descriptors.　And　then　we　estimated　the　homography　matrix　with　the　selected　sample.　The　homography　matrix　is　used　to　eliminate　the　characteristics　of　mismatch　Furthermore,　a　binary　environment　dictionary　is　created　for　loop　detection　and　the　experimental　results　demonstrate　that　this　method　improves　the　speed　of　loading　time　of　the　dictionary　and　the　accuracy　of　SLAM.