Abstract: Existing methods to quantitatively evaluate the performance of image-based 3D reconstruction are generally based on ground-truth 3D models acquired by 3D scanner, with disadvantages of high cost, low efficiency, and inapplicability to large-scale objects or scenes.To solve this problem, propose an evaluation method based on the invariance of image's phase information.Firstly, get the 2D projection image of the 3D reconstructed model under certain viewpoint.Then, extract edges of this projection image as well as the image of realistic object under the similar viewpoint.Subsequently, compute the proposed phase moment invariants for the two edge images and construct feature vectors.Finally, assess the reconstructed result through comparing the distance between the two feature vectors.Experiments validate the effectiveness and applicability of this method for assessing 3D reconstructed surface model.