This　paper　provides　a　novel　content-based　image　retrieval　algorithm　based　on　ROI　(Region　Of　Interest)　by　using　SIFT　(Scale　Invariant　Feature　Transform)　feature　matching.　SIFT　descriptors,　which　are　invariant　to　image　scaling　and　transformation　and　rotation,　and　partially　invariant　to　illumination　changes　and　affine,　present　the　local　features　of　an　image.　Therefore,　feature　keypoints　can　be　extracted　more　accurately　by　using　SIFT　from　user-defined　ROI　of　an　image　than　color,　texture,　shape　and　spatial　relations　feature.　To　decrease　unavailable　features　matching,　a　dynamic　probability　function　replaces　the　original　fixed　value　to　determine　the　similarity　distance　of　ROI　and　database　from　training　images.　These　are　the　kernel　of　content-based　image　retrieval.　The　experimental　results　show　that　this　method　improves　the　stability　and　precision　of　image　retrieval.