Recent　progress　in　enhancing　low-light　images　using　deep　learning　techniques　has　been　significant.　Despite　these　strides,　the　current　methodologies　often　treat　the　entire　image　as　a　homogeneous　entity,　neglecting　the　incorporation　of　semantic　information　from　distinct　regions.　Consequently,　this　approach　may　lead　to　a　network　deviating　from　the　original　color　characteristics　of　specific　regions.　To　address　this　challenge,　we　present　a　new　two-stage　model　for　low-light　image　enhancement　called　the　semantic-guided　brightness　curve　estimation　network　(SBC-Net).　In　our　proposed　SBC-Net,　we　use　the　method　of　brightness　curve　to　enhance　lighting.　A　discrete　brightness　curve　is　defined　to　satisfy　the　monotonicity　of　the　curve　through　its　first-order　derivative.　SBC-NET　first　performs　image　segmentation　using　the　recently　proposed　segment　anything　model　to　calculate　discrete　brightness　curves　for　different　semantic　regions.　Then,　a　semantic-guided　long-range　and　short-range　denoising　model　is　applied　to　perform　detail　restoration　on　the　brightness-enhanced　image,　using　different　models　for　detail　recovery　in　different　semantic　areas　based　on　the　level　of　noise.　Comprehensive　experimental　results　across　various　datasets　affirm　the　superior　performance　of　SBC-Net,　demonstrating　excellence　in　both　quantitative　metrics　and　visual　quality.　Project　page:　https://github.com/LambChuckEye/Semantic-Guided-Brightness-Curve-Estimation-for-Low-Light-Image-Enhancement.