As　existing　CBIR　systems　are　difficult　to　bridge　the　semantic　gap,　a　new　region　of　interest　(ROI)　detection　method　is　proposed　based　on　gaze　tracking.　Eye-gaze　conveys　a　great　deal　of　information　including　the　underlying　mental　process　of　the　subject.　The　user　interests　are　real-time　obtained　by　tracking　gaze　points,　then　the　ROI　is　detected.　A　real-time　and　low　complexity　gaze　tracking　method　is　proposed　and　the　interest　measure　functions　are　defined.　A　platform　for　gaze　tracking　based　ROI　detection　is　designed　and　implemented.　Experimental　results　show　that　the　ROI　can　be　detected　by　the　method,　and　the　semantic　gap　is　bridged.