Thigh muscle morphology and composition quantified from MR images are potential imaging biomarkers for diseases such as osteoarthritis and sarcopenia. MR thigh muscle segmentation is an important step in quantifying both muscle morphology and composition. Unfortunately, the thigh muscle groups are tightly bundled together, making them very hard to segment due to a lack of clear boundaries between different muscles. We proposed a novel geometric flow based semi-automatic scheme to effectively segment them. We combined reproducible kernel Hilbert space edge descriptor and geodesic distance maps from a set of markers and anti-markers to define the force for the geometric flow.