In　this　paper,　a　secure　image　retrieval　method　for　cloud　computing　is　proposed,　which　can　guarantee　the　security　of　the　image　content　while　not　sacrificing　the　retrieval　performance.　The　method　is　implemented　based　on　CBIR　(content-based　image　retrieval)　framework.　To　perform　the　secure　image　retrieval,　three　kinds　of　low-level　image　features　are　firstly　extracted　from　images,　including　color,　texture　and　shape　features.　Locality　preserving　projections　(LPP)　is　then　applied　to　reduce　the　dimension　of　these　image　features.　Secondly,　image　features　are　protected　by　the　Paillier　homomorphic　encryption　algorithm　according　to　its　homomorphic　characteristics.　Finally,　similarity　measurement　is　directly　conducted　between　the　encrypted　features,　and　the　top　12　similar　images　are　treated　as　the　retrieval　results.　The　security　and　communication　cost　analysis　of　the　proposed　method　are　analyzed.　Experimental　results　demonstrate　that　the　proposed　method　can　achieve　the　consistent　retrieval　results　with　the　conventional　CBIR　method　in　the　plaintext　domain　while　providing　adequate　security.　©　2014　IEEE.