Aiming at the drawback of the classical total variation (TV) de-noising model in which the staircase effect is often produced, an image de-noising model based on improved second order total generalized variation (TGV) is proposed. In the new model, the image texture information extracted by Kirsch edge detection operator is used to introduce an edge indicator function to guide diffusion in the regularization term of second order TGV. The experiment shows that compared with the classical TV de-noising model and the second-order TGV de-noising model, the new model has obvious improvement in both visual effect, peak signal-to-noise ratio (PSNR) and mean square error (MSE), which can remove the noise effectively while protecting image edge information and fine texture structure information adaptively.