The accurate segmentation of ship waterline areas is critical for ship draft surveys and maritime measurement. However, existing methods are often hindered by environmental variations, low contrast between water and hull, reflections, and shadows, leading to missed or inaccurate detections. To address these challenges, this paper proposes SCTLine, a dual-branch network integrating local detail extraction and global context modeling. By simultaneously preserving edge details and overall continuity, SCTLine enhances the perception of slender and discontinuous waterline structures. The outputs from both branches are spatially aligned and semantically fused via a lightweight decoder, yielding high-resolution, pixel-level segmentation results. Extensive experiments on a self-built dataset of 6,000 high-resolution ship images demonstrate that SCTLine achieves superior segmentation accuracy and boundary quality compared to mainstream methods, while maintaining high inference efficiency.