Aiming　at　the　problem　that　frequent　occurrences　of　ocular　artifacts　seriously　interfere　with　the　electroencephalogram(EEG)　interpretation　and　analysis,　a　novel　technique　to　eliminate　ocular　artifacts　from　EEG　signals　in　real-time　is　proposed.　The　independent　component　analysis(ICA)　is　employed　to　decompose　EEG　signals,　and　these　independent　components　features　of　topography　and　power　spectral　density　are　extracted.　Specifically,　a　template-based　isometric　mapping(Isomap)　algorithm　is　adopted　to　reduce　the　feature　dimensionality.　The　low-dimensional　feature　samples　are　fed　to　a　classifier　to　identify　ocular　artifacts　components.　The　classification　performances　of　several　typical　classifiers　show　that　the　template-based　Isomap　algorithm　with　the　nearest　neighbor　classifier　performs　best.　The　experimental　results　demonstrate　the　efficiency　for　removing　ocular　artifacts　with　little　distortion　of　underlying　brain　signals.