In this paper we propose a contour extraction method based on the active contour model, which uses the GVF snake to obtain the initial segments for a contour; and then a minimal path method for the refinement stage, to obtain an accurate and more robust result. By employing the minimal path method to find missing segments between pairs of nodes defined on the contour obtained by the GVF snake, our algorithm is able to detect deep concave parts of object boundary, and works well even when the snake initialization is not very good.