Abstract: Extracting continuous and high-quality face contour is an important preprocessing step of many high-level image applications. However, existing approaches are not dedicated algorithms for face contour extraction. In this paper, we find that it is feasible and economical to refine sparse landmarks to obtain a continuous and high-quality face contour, based on the fact that the landmarks can provide us with an initial face curve. The method proposed is composed of three steps: Firstly, e sample small and overlapped squares along the initial curve to cut the target face contour into small segments. We assume that any local segment of a face contour resembles a parabola, which is more reasonable than assuming the whole face contour as a parabola. Then locally in each square, we propose a parabola-constrained seam cutting algorithm to directly identify the face contour segment in the square. Finally, we propose a global seam integrating algorithm that extracts the target face contour from all the local segments. This cross-validation mechanism guarantees the success of the extraction of the final face contour. Experiments on two modern face datasets, i.e., LFPW and HELEN, demonstrate that our method significantly improves the results of the state-of-the-art face alignment approaches.