In　order　to　solve　the　problem　of　content-based　image　retrieval,　this　paper　defines　a　new　concept　of　Information　Entropy　Shift,　and　proposes　a　novel　image　retrieval　algorithm　based　on　it.　The　Information　Entropy　Shift　is　defined　as　the　changes　of　Information　Entropy　in　one　direction;　it　not　only　describes　the　statistical　feature　of　an　image,　but　also　reflects　the　spatial　Information　distribution.　On　the　basis　of　this　concept,　this　paper　presents　a　new　dynamic　margin　matching　algorithm.　In　this　algorithm,　the　matching　problem　of　information　entropy　shift　sequence　is　transformed　to　the　matching　of　the　characteristic　curve　of　a　binary　function;　the　trend　of　curve　changes　is　used　as　the　feature　to　match.　Besides,　the　template　matching　algorithm　in　audio　processing　is　referred　to　deal　with　the　problem　too.　Experimental　results　show　that　the　new　algorithm　has　the　advantage　of　high　efficiency　and　is　robust　to　image　resizing　and　translation.