This　paper　presents　a　new　measure　method　of　stereo　matching　algorithm,　which　is　based　on　image　structure　and　pixel＇s　feature,　named　CGDSM.　The　matching　cost　calculate　step　consist　of　three　similarity　measure　factors:　pixel　gradient　similarity　measure　factor,　pixel　feature　similarity　measure　factor　and　image　structural　similarity　measure　factor.　At　the　cost　aggregate　step　we　use　adaptive　window　to　aggregate　which　is　based　on　Canny　edge　information.　At　the　disparity　calculate　step　we　use　WTA(winner　takes　all)　algorithm　to　find　the　optimal　disparity.　At　the　disparity　refine　step　we　use　LRC(left-right　consistency)　method　to　refine　disparity.　The　new　algorithm　contain　more　similarity　measure　function　from　image　gradient　information　to　color　domain　feature　information　and　then　to　the　spatial　structural　information.　Experiment　shows　that　this　algorithm　enhance　the　accuracy　of　stereo　matching　algorithm.