A method of image interpolation with non-local feature directions is proposed.This method respectes the smooth of the contour profile of interpolated image and retrains blur edges.The eigenvector of the non-local Hessian matrix is considered as the image featrue direction.The diffusion of image energy functional is controlled by the eigenvalue of image local Hessian along the direction.It overcomes the local limit of gradient pointing image feature and drives image energy functional to diffuse along corrected direction.Thus the blur of image feature is avoids.Numerical experiments on real images show that images interpolated by the proposed method have better interpolated edges and are almost artifact-free.