In　the　research　of　the　self-aligned　dicing　technology,　due　to　tens　of　thousands　of　types　of　wafer　image　samples,　there　are　diverse　types　of　street　in　wafer　image.　In　order　to　properly　identify　the　paddle　vat　in　the　street　and　get　center　line　of　the　paddle　vat　for　cutting　by　using　binary　segmentation,　it＇s　a　crucial　step　for　binarization　of　wafer　images.　To　take　into　account　the　diversity　of　the　wafer　images,　it　is　quite　difficult　to　find　a　good　threshold.　In　this　paper,　we　propose　a　binarization　approach　based　on　the　density　histogram　to　locate　the　center　line　of　paddle　vat　of　the　wafer　image.　The　histogram　of　a　wafer　image　is　smoothed　by　using　a　quadratic　exponential　function　and　then　an　appropriate　threshold　is　calculated.　The　experiments　show　that　the　new　method　is　very　effective　for　the　wafer　image　processing.