Accurate image segmentation remains a cornerstone challenge in computer vision, particularly under open-set conditions where object variability and scene complexity hinder generalization. To address these limitations, we propose a novel visual-based methodology entitled Visual-Based Space Debris Segmentation Using an Enhanced Segment Anything Framework. This approach synergistically integrates an optimized clause-aware prompt mechanism derived from Grounding DINO with a structurally refined version of the Segment Anything Model (SAM). By embedding hierarchical non-maximum suppression and adaptive region purification through connected component filtration, we substantially augment segmentation fidelity. Furthermore, we incorporate ViT-Matte, a vision transformer-based trimap enhancement module, to improve boundary localization and reduce aliasing in edge delineation. Extensive validation on the COCO2017 benchmark reveals that our framework elevates Mean Pixel Accuracy by 6.04%, culminating at 24.74%, thereby substantiating its efficacy in foreground-background discrimination under visually ambiguous scenarios such as orbital debris fields.