Low　dose　computed　tomography　(LDCT)　can　reduce　the　radiation　hazard　to　patients　effectively.　However,　mottle　noise　and　streak　artifacts　often　lower　and　degenerate　the　quality　of　the　LDCT　image　in　the　process　of　reconstruction.　This　article　presents　a　two-step　denoising　method,　which　exploits　the　morphological　component　analysis　(MCA)　and　non-local　means　(NLM),　for　removing　the　noise　and　artifacts　in　LDCT　image.　In　the　first　step,　the　MCA-based　image　separation　is　performed　with　the　proposed　dictionary.　The　dictionary　is　firstly　established　from　the　learning　procedure　from　the　preprocessed　images,　and　then　modified　by　using　gradient　activity　measure.　Consequently,　the　streak　artifacts　are　removed　from　LDCT　image.　In　the　second　step,　the　NLM　method　is　adopted　to　further　remove　the　mottle　noise　in　the　residual　image.　Experimental　results　from　both　simulated　phantom　and　real　clinical　data　demonstrate　that　compared　with　several　related　methods,　the　proposed　method　shows　superior　performance　in　both　noise/artifacts　removal　and　structure　preservation.