As　for　the　huge　gap　between　the　low-level　image　features　and　high-level　semantics,　content-based　image　retrieval　still　could　not　receive　a　satisfying　result　by　now.　Since　the　special　request　of　the　relevance　feedback,　making　full　use　of　the　rare　number　of　labeled　data　and　numerous　unlabeled　data　is　an　ideal　way.　Because　ELM　has　excellent　classification　accuracy　and　processing　time,　and　high　accuracy　and　fast　speed　are　the　key　factors　to　evaluate　the　relevance　feedback　performances.　In　this　paper,　we　proposed　an　Extreme　learning　Machine　with　Gaussian　kernel　Based　Relevance　Feedback　scheme　for　image　retrieval,　to　overcome　the　above　limitations,　our　method　uses　three　component　classifiers　to　form　a　strong　learner　by　learning　different　features　extracted　from　the　hand-marking　data,　then　we　use　it　to　label　the　image　database　automatically.　From　the　experiments　we　can　see　the　use　of　the　ELM　with　kernel　have　high　classification　accuracy,　the　processing　time　get　largely　decreased　at　the　same　time.　Thus,　it　improves　the　efficiency　of　entire　relevance　feedback　system.　The　experiments　results　show　that　the　proposed　algorithm　is　significantly　effective.