We propose an original method for vectorizing an image or zooming it at an\narbitrary scale. The core of our method relies on the resolution of a geometric\nvariational model and therefore offers theoretic guarantees. More precisely, it\nassociates a total variation energy to every valid triangulation of the image\npixels. Its minimization induces a trian-gulation that reflects image\ngradients. We then exploit this triangulation to precisely locate\ndiscontinuities, which can then simply be vectorized or zoomed. This new\napproach works on arbitrary images without any learning phase. It is\nparticularly appealing for processing images with low quantization like pixel\nart and can be used for depixelizing such images. The method can be evaluated\nwith an online demonstrator, where users can reproduce results presented here\nor upload their own images.\n