As　the　reality　that　human　beings　usually　pay　more　attention　to　areas　of　interest,　visual　attention　model　is　a　feasible　method　to　find　Regions　of　interest　(ROIs)　and　measure　the　interest　of　a　region.　However,　it　is　required　to　decompress　image　data　completely.　A　visual　attention　model　based　ROIs　detection　in　compressed　domain　is　proposed　in　this　paper,　which　can　compute　visual　attention　model　with　partially　decompression.　This　method　includes:　(1)　Visual　saliency　map　computation;　(2)　Focus　of　attention　(FOA)　selection　and　shift;　(3)　ROIs　detection.　The　experimental　results　show　the　proposed　method　performs　well　on　the　speed/accuracy　of　ROIs　detection　and　interest　measurement.