In　this　paper,　three-dimensional　data　reconstruction　with　a　small　scale　based　on　micro　stereo-vision　is　discussed.　An　inversion　model　of　stereo-vision　that　connects　the　object　space　and　the　image　space　is　derived　to　obtain　three-dimensional　coordinates　of　object　point　in　the　object　space　from　two-dimensional　image　coordinates　of　a　pair　of　image　points　in　the　image　space.　An　algorithm　based　on　multi　similarity　measurements　and　multi-matching　areas　is　proposed　for　stereo　matching　of　color　images,　which　considers　the　impacts　of　similarity　measurements　and　matching　areas　fully.　In　this　algorithm,　the　samples　of　matching　output　are　acquired　by　the　statistics　of　multi　similarity　measurements　and　multi-matching　areas.　On　the　other　hand,　the　false　matching　results　and　the　wrong　matching　results　are　partially　filtered　by　performing　the　mean-square　error　statistics　of　matching　output　samples.　Furthermore,　a　sub-pixel　matching　accuracy　is　expected　by　calculating　the　average　values　of　two　new　samples　derived　from　the　mean-square　error　statistics.　In　the　software　platform　based　on　VC++　and　OpenGL,　computer-aided　design　is　designed　for　filter　of　abnormal　data.