Image　saliency　detection,　which　is　aiming　at　locating　meaningful　objects　in　image　and　extracting　the　corresponding　saliency　map,　has　been　attracting　considerable　attention.　Most　of　existing　saliency　detection　algorithms　are　dealing　with　uncompressed　images,　while　numerous　images　are　distributed　and　stored　in　JPEG　format　over　Internet.　Thus,　this　paper　proposes　a　novel　image　saliency　detection　method　dealing　with　JPEG　images,　which　directly　extracts　3　image　low-level　features　(　intensity,　texture　and　color)　from　JPEG　DCT　blocks　and　then　evaluates　image　saliency　by　calculating　the　shortest　path　from　each　internal　point　to　the　background.　We　compare　our　method　with　some　state-of-the-art　methods　on　the　publicly　available　dataset　MSRA-1000.　Experimental　results　show　that　our　method　exhibits　better　performance　in　terms　of　all　four　evaluations　than　some　state-of-the-art　methods　in　our　experiments.　The　final　saliency　maps　indicate　that　our　method　can　also　produce　satisfied　saliency　maps　directly　in　compressed　domain.