A level-set-based method using a radial basis functions (RBFs) based level set interpolation with an ellipse constraint is presented for image contour extraction. In the present method, the initial distance function embedded in the ellipse-constrained RBFs is interpolated using a coarse grid. The deformation of the level set function (LSF) is considered as an update of the RBFs’ coefficients by solving an ordinary differential equation (ODE) and non-convex constrained quadratic programming (QCQP). A semi-definite relaxation approach is proposed to solve the non-convex QCQP problem. The proposed level set evolving scheme, which does not need initialization and re-initialization, is efficient and does not suffer from self-flattening. The objects with extremely complex shapes can be exactly fitted with a coarse grid of RBFs’ centers and the image extraction is less sensitive to the distribution of the objects in the image domain.