This　work　proposes　a　novel　scheme　for　encrypted　JPEG　image　retrieval,　which　includes　image　encryption　and　unsupervised/supervised　retrieval　phases.　Using　this　scheme,　the　encrypted　images　are　produced　by　permuting　DCT　coefficients,　and　transmitted　to　a　database　server.　With　an　encrypted　query　image,　although　the　server　does　not　know　the　plaintext　content,　he　may　get　the　histogram　at　each　frequency　position.　After　calculating　the　distances　between　the　histograms　of　encrypted　query　image　and　database　image,　the　server　can　return　the　encrypted　images　with　plaintext　content　similar　to　the　query　image　according　to　integrated　distances.　If　a　training　image　set　is　available,　the　retrieval　results　can　be　also　determined　by　conditional　probabilities　calculated　from　a　supervised　mechanism.