Speech　endpoint　detection　is　an　important　step　in　speech　preprocessing.　Its　accuracy　has　an　important　impact　on　the　speech　synthesis　and　recognition　systems.　Based　on　the　feature　of　formant-consonance　energy,　a　scheme　of　speech　endpoint　detection　is　presented　by　taking　advantage　of　image　processing　technology.　This　algorithm　firstly　gets　rid　of　the　periodic　interference.　Then　it　segments　the　spectrogram　image　after　de-noising.　Finally　it　derives　the　feature　of　formant　and　speech　endpoint　by　using　first　order　Gaussian　differential　filter.　The　experimental　results　indicate　that　it　can　get　good　quantitative　results　in　the　experiments　within　the　environment　of　different　SNR　white　noise　and　outburst　noise.